segmentation/summary.md

Experiment Suite Summary

Metric Notes

• Main comparison metric: test_miou_present.
• Complexity is reported as gflops (2 FLOPs per MAC convention).
• params_m/gflops/latency are inference-time model complexity numbers (independent of whether some modules were frozen during training).
• MCPNet upstream default schedule: 80k iters (configs/_base_/schedules/schedule_80k.py).
• PPMambaSeg upstream defaults vary by dataset; PPMamba configs are typically 40-50 epochs (e.g., LoveDA=40, Vaihingen=50).

Ranking (chronological order)

status	model	backbone	loss	venue	params_m	gflops	latency_ms_1x3x512x512	peak_vram_gb	test_miou_present	test_macro_f1_present	test_oa_fg	test_miou	best_val_miou	best_val_miou_present
-	--- GENERAL SEGMENTATION MODELS ---	-	-	-	-	-	-	-	-	-	-	-	-	-
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	ECCV2018	29.3108	75.9139	4.4695	0.2086	0.3895	0.5260	0.7189	0.2782	0.3292	0.3545
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	ECCV2018	29.3108	75.9139	4.4695	0.2086	0.3766	0.5193	0.6840	0.2690	0.3482	0.3750
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	ECCV2018	29.3108	75.9139	4.4695	0.2086	0.3735	0.5068	0.7048	0.2668	0.3325	0.3581
completed	DeepLabV3+	ResNet-50	ce+dice	ECCV2018	26.6809	73.5922	3.0034	0.2300	0.3662	0.4897	0.7092	0.2817	0.3379	0.3379
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	ECCV2018	26.6809	73.5922	3.0034	0.2300	0.3438	0.4755	0.6951	0.2644	0.3427	0.3427
completed	DeepLabV3+	ResNet-50	focal+dice	ECCV2018	26.6809	73.5922	3.0034	0.2300	0.3515	0.4819	0.7000	0.2929	0.3415	0.3461
completed	UPerNet	Swin-Tiny	ce+dice	ECCV2018	59.8371	472.1168	22.6344	0.5306	0.3371	0.4651	0.6729	0.2593	0.2873	0.2873
completed	OCRNet	HRNet-W48	ce+dice	ECCV2020	70.3653	325.3542	61.4944	0.5052	0.2722	0.3954	0.5735	0.2268	0.2954	0.2954
completed	SegFormer	MiT-B2	ce+dice	NeurIPS2021	27.3574	121.9349	8.2788	0.5439	0.3222	0.4594	0.6484	0.2301	0.3231	0.3480
completed	SegFormer	MiT-B2	weighted_ce+dice	NeurIPS2021	27.3574	121.9349	8.2788	0.5439	0.4010	0.5297	0.7163	0.3084	0.3423	0.3423
completed	SegFormer	MiT-B2	focal+dice	NeurIPS2021	27.3574	121.9349	8.2788	0.5439	0.3332	0.4760	0.6385	0.2380	0.3250	0.3501
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	CVPR2022	44.0064	133.2907	17.4630	0.4213	0.2985	0.4285	0.6561	0.2985	0.3033	0.3033
completed	SegNeXt	MSCAN-Tiny	ce+dice	NeurIPS2022	4.2285	12.6449	9.2612	0.1673	0.2682	0.3884	0.6065	0.1916	0.2895	0.2896
completed	Afformer	AFFormer-Base	ce+dice	AAAI2023	2.9690	8.5730	7.4704	0.1691	0.3047	0.4362	0.6389	0.2176	0.2928	0.3153
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	ICCV2023	15.2802	18.3156	6.4212	0.1213	0.3799	0.5065	0.7258	0.2713	0.3444	0.3444
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	ICCV2023	15.2802	18.3156	6.4212	0.1213	0.3790	0.5181	0.7154	0.2707	0.3565	0.3840
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	ICCV2023	15.2802	18.3156	6.4212	0.1213	0.3693	0.5086	0.6781	0.2638	0.3520	0.3790
completed	SeaFormer	SeaFormer-Base	ce+dice	ICLR2023	8.5838	3.4741	12.4666	0.1700	0.3117	0.4408	0.6392	0.2398	0.3116	0.3116
completed	CGRSeg	EfficientFormerV2-B	ce+dice	ECCV2024	19.0799	7.5003	14.4632	0.2569	0.2679	0.3961	0.5844	0.1913	0.3054	0.3054
completed	PEM	ResNet-50	set_matching_ce+mask+dice	CVPR2024	35.5313	60.6003	11.5152	0.3881	0.2789	0.4011	0.6502	0.2789	0.2549	0.2549
-	--- REMOTE-SENSING-SPECIFIC METHODS ---	-	-	-	-	-	-	-	-	-	-	-	-	-
completed	FarSeg	ResNet-50	ce (native)	CVPR2020	31.3698	94.1161	3.9178	0.2414	0.3564	0.4726	0.7130	0.2970	0.2989	0.2989
completed	FarSeg	ResNet-50	ce+dice	CVPR2020	31.3698	94.1161	3.9178	0.2414	0.3717	0.4934	0.6893	0.3098	0.3111	0.3111
completed	FarSeg	ResNet-50	weighted_ce+dice	CVPR2020	31.3698	94.1161	3.9178	0.2414	0.3642	0.5066	0.6636	0.2601	0.3073	0.3310
completed	FarSeg	ResNet-50	focal+dice	CVPR2020	31.3698	94.1161	3.9178	0.2414	0.3744	0.4899	0.7342	0.3120	0.3286	0.3286
completed	BANet	ResT-Lite	ce+dice	RS2021	12.8608	31.3805	4.8148	0.1029	0.2926	0.4147	0.6535	0.2250	0.2779	0.2992
completed	ABCNet	ResNet-18	ce+dice+aux_ce	ISPRSJPRS2021	13.9645	32.3860	2.8119	0.1004	0.3145	0.4302	0.6831	0.2621	0.3070	0.3070
completed	MANet	ResNet-50	ce+dice	TGRS2022	35.8629	109.6158	4.7940	0.3940	0.3711	0.4922	0.6862	0.3093	0.3147	0.3147
completed	MANet	ResNet-50	weighted_ce+dice	TGRS2022	35.8629	109.6158	4.7940	0.3940	0.3828	0.5228	0.6759	0.2734	0.3079	0.3316
completed	MANet	ResNet-50	focal+dice	TGRS2022	35.8629	109.6158	4.7940	0.3940	0.3999	0.5431	0.6848	0.2856	0.3015	0.3246
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	ISPRSJPRS2022	11.7259	23.5509	3.3975	0.0876	0.3941	0.5152	0.7276	0.3284	0.3388	0.3388
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	ISPRSJPRS2022	11.7259	23.5509	3.3975	0.0876	0.3566	0.4931	0.6811	0.2743	0.3275	0.3275
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	ISPRSJPRS2022	11.7259	23.5509	3.3975	0.0876	0.3765	0.4989	0.7182	0.3137	0.3314	0.3314
completed	DC-Swin	Swin-Small	ce+dice	TGRS2022	66.9503	144.3925	12.3804	0.3561	0.2971	0.4173	0.6584	0.2476	0.2884	0.2884
completed	A2FPN	ResNet-18	ce+dice	IJRS2022	12.1620	27.1366	2.1229	0.2150	0.3688	0.4834	0.7335	0.3073	0.3085	0.3085
completed	A2FPN	ResNet-18	weighted_ce+dice	IJRS2022	12.1620	27.1366	2.1229	0.2150	0.3720	0.5107	0.7094	0.2657	0.2959	0.3187
completed	A2FPN	ResNet-18	focal+dice	IJRS2022	12.1620	27.1366	2.1229	0.2150	0.3363	0.4602	0.6659	0.2802	0.3027	0.3027
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	ICASSP2023	30.9157	99.2253	6.0530	0.2298	0.3108	0.4081	0.7474	0.2590	0.2951	0.2951
completed	FarSeg++	MiT-B2	ce (native)	TGRS2023	32.5566	95.0793	8.7746	0.2784	0.3062	0.4358	0.6669	0.2187	0.3049	0.3284
completed	SACANet	HRNet-W32	ce+aux_ce (native)	ICME2023	30.2704	115.9042	19.0179	0.3073	0.3294	0.4557	0.6573	0.2534	0.2985	0.3215
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	GRSL2024	39.1269	395.3173	20.6364	0.4263	0.3147	0.4398	0.6785	0.2421	0.2772	0.2772
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	GRSL2025	21.7049	45.9905	11.2756	0.3103	0.3520	0.4780	0.6683	0.2934	0.3362	0.3362
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	GRSL2025	21.7049	45.9905	11.2756	0.3103	0.3854	0.5298	0.6816	0.2753	0.3466	0.3466
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	GRSL2025	21.7049	45.9905	11.2756	0.3103	0.3897	0.5100	0.7103	0.3248	0.3411	0.3411
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	GRSL2024	43.3254	78.5912	11.6012	0.4624	0.2385	0.3080	0.7257	0.1987	0.1559	0.1559
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	GRSL2024	43.3254	78.5912	11.6012	0.4624	0.3068	0.4280	0.6519	0.2556	0.2313	0.2313
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	GRSL2024	43.3254	78.5912	11.6012	0.4624	0.2399	0.3125	0.7251	0.1999	0.1910	0.1910
completed	PyramidMamba	Swin-Base	ce+dice	JAG2025	125.1077	217.7548	13.7581	0.6582	0.3985	0.5360	0.6833	0.2847	0.3703	0.3703
completed	PyramidMamba	Swin-Base	weighted_ce+dice	JAG2025	125.1077	217.7548	13.7581	0.6582	0.4414	0.5864	0.6967	0.3153	0.3830	0.4125
completed	PyramidMamba	Swin-Base	focal+dice	JAG2025	125.1077	217.7548	13.7581	0.6582	0.3961	0.5304	0.6699	0.2830	0.3685	0.3685
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	TGRS2025	25.1927	74.3696	17.1870	0.2225	0.2264	0.3066	0.6353	0.2264	0.2264	0.2264
completed	MF-Mamba	HRNet-W18	ce+dice	TGRS2025	11.2729	38.9439	20.5415	0.1326	0.3001	0.4242	0.6376	0.2501	0.3039	0.3039
completed	MCPNet	ResNet-50	ce+dice	TGRS2025	45.1516	110.9866	6.8757	0.3528	0.3056	0.4267	0.6680	0.2183	0.3051	0.3051
completed	MCPNet	ResNet-50	weighted_ce+dice	TGRS2025	45.1516	110.9866	6.8757	0.3528	0.3193	0.4552	0.6405	0.2281	0.2954	0.3181
completed	MCPNet	ResNet-50	focal+dice	TGRS2025	45.1516	110.9866	6.8757	0.3528	0.3233	0.4448	0.6898	0.2487	0.3027	0.3027
-	--- METHODS RELATED TO VISION FOUNDATION MODELS ---	-	-	-	-	-	-	-	-	-	-	-	-	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	NeurIPS2023	97.8294	983.1302	74.3363	2.7571	0.2485	0.3558	0.6390	0.1775	0.2067	0.2226
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	NeurIPS2023	97.8294	983.1302	74.3363	2.7571	0.2538	0.3711	0.6150	0.1813	0.2077	0.2237
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	NeurIPS2023	97.8294	983.1302	74.3363	2.7571	0.2503	0.3561	0.6539	0.1925	0.2087	0.2248
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	TGRS2024	13.9645	32.3516	2.5705	0.1014	0.2964	0.4098	0.6573	0.2470	0.3104	0.3104
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	TGRS2024	30.0727	66.1392	6.2155	0.3345	0.2916	0.4084	0.6598	0.2243	0.2909	0.2909
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	TGRS2024	96.1376	51.0381	14.6563	0.4374	0.2922	0.4094	0.6871	0.2435	0.2859	0.2859
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	TGRS2024	11.6880	23.5177	3.3229	0.0874	0.3241	0.4452	0.6839	0.2700	0.2971	0.2971
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	ICLR2025	83.8976	191.8167	10.4857	0.5898	0.2422	0.3510	0.6193	0.1863	0.2082	0.2242
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	ICLR2025	83.8976	191.8167	10.4857	0.5898	0.2207	0.3235	0.5938	0.1577	0.2160	0.2326
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	ICLR2025	83.8976	191.8167	10.4857	0.5898	0.2351	0.3438	0.6158	0.1809	0.2124	0.2288
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	ICLR2025	83.8976	191.8167	10.4857	0.5898	0.2885	0.4089	0.6562	0.2405	0.2903	0.2903
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	ICLR2025	83.8976	191.8167	10.4857	0.5898	0.2875	0.4058	0.6769	0.2054	0.2723	0.2933
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	ICLR2025	83.8976	191.8167	10.4857	0.5898	0.2980	0.4155	0.6870	0.2483	0.2906	0.2906
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	RS2025	98.5875	247.0546	15.3571	0.6103	0.3263	0.4472	0.6978	0.2510	0.2959	0.2959
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	RS2025	98.5875	247.0546	15.3571	0.6103	0.3696	0.5085	0.6978	0.2640	0.3128	0.3333
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	RS2025	98.5875	247.0546	15.3571	0.6103	0.3450	0.4642	0.7430	0.2654	0.2841	0.2841
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	TGRS2025	12.1620	27.1366	3.2010	0.2150	0.3702	0.4848	0.7338	0.3085	0.3094	0.3094
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	TGRS2025	12.1620	27.1366	3.3198	0.2150	0.3374	0.4613	0.6663	0.2812	0.3035	0.3035
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	TGRS2025	12.1620	27.1366	3.2376	0.2150	0.3745	0.5139	0.7098	0.2675	0.2984	0.3214
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	TGRS2025	13.9645	32.3860	20.9519	0.1004	0.3154	0.4311	0.6835	0.2629	0.3078	0.3078
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	TGRS2025	12.8608	31.3805	7.7093	0.1029	0.2937	0.4161	0.6536	0.2259	0.2791	0.3006
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	TGRS2025	35.8629	109.6158	7.7557	0.3940	0.3604	0.4820	0.6775	0.3004	0.3162	0.3162
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	TGRS2025	35.8629	109.6158	7.0723	0.3940	0.4032	0.5467	0.6849	0.2880	0.1723	0.3015
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	TGRS2025	35.8629	109.6158	8.6225	0.3940	0.3839	0.5238	0.6763	0.2742	0.3085	0.3322
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	TGRS2025	11.7259	23.5509	4.9343	0.0876	0.3958	0.5167	0.7279	0.3298	0.3399	0.3399
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	TGRS2025	11.7259	23.5509	4.7844	0.0876	0.3873	0.5091	0.7195	0.3228	0.3406	0.3406
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	TGRS2025	11.7259	23.5509	5.0644	0.0876	0.3578	0.4943	0.6816	0.2752	0.3286	0.3286

Training / Validation Summary

status	model	backbone	loss	venue	best_val_miou_epoch	best_val_miou	best_val_miou_present_epoch	best_val_miou_present	val_bestckpt_macro_f1_present	val_bestckpt_oa_fg
-	--- GENERAL SEGMENTATION MODELS ---	-	-	-	-	-	-	-	-	-
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	ECCV2018	31	0.3292	31	0.3545	0.4635	0.7756
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	ECCV2018	29	0.3482	29	0.3750	0.4984	0.7134
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	ECCV2018	62	0.3325	62	0.3581	0.4709	0.7464
completed	DeepLabV3+	ResNet-50	ce+dice	ECCV2018	12	0.3379	12	0.3379	0.4502	0.7204
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	ECCV2018	23	0.3427	23	0.3427	0.4627	0.7224
completed	DeepLabV3+	ResNet-50	focal+dice	ECCV2018	65	0.3415	12	0.3461	0.4657	0.7042
completed	UPerNet	Swin-Tiny	ce+dice	ECCV2018	31	0.2873	31	0.2873	0.3876	0.6733
completed	OCRNet	HRNet-W48	ce+dice	ECCV2020	47	0.2954	47	0.2954	0.4008	0.6782
completed	SegFormer	MiT-B2	ce+dice	NeurIPS2021	32	0.3231	32	0.3480	0.4556	0.7732
completed	SegFormer	MiT-B2	weighted_ce+dice	NeurIPS2021	2	0.3423	60	0.3423	0.4523	0.7439
completed	SegFormer	MiT-B2	focal+dice	NeurIPS2021	17	0.3250	17	0.3501	0.4640	0.7633
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	CVPR2022	40	0.3033	40	0.3033	0.4128	0.7111
completed	SegNeXt	MSCAN-Tiny	ce+dice	NeurIPS2022	75	0.2895	75	0.2896	0.3805	0.7429
completed	Afformer	AFFormer-Base	ce+dice	AAAI2023	76	0.2928	76	0.3153	0.4038	0.7610
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	ICCV2023	13	0.3444	64	0.3444	0.4580	0.7550
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	ICCV2023	68	0.3565	68	0.3840	0.5164	0.7507
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	ICCV2023	50	0.3520	50	0.3790	0.5034	0.7380
completed	SeaFormer	SeaFormer-Base	ce+dice	ICLR2023	37	0.3116	37	0.3116	0.4020	0.7642
completed	CGRSeg	EfficientFormerV2-B	ce+dice	ECCV2024	56	0.3054	56	0.3054	0.3963	0.7577
completed	PEM	ResNet-50	set_matching_ce+mask+dice	CVPR2024	57	0.2549	57	0.2549	0.3483	0.6476
-	--- REMOTE-SENSING-SPECIFIC METHODS ---	-	-	-	-	-	-	-	-	-
completed	FarSeg	ResNet-50	ce (native)	CVPR2020	10	0.2989	10	0.2989	0.4017	0.6666
completed	FarSeg	ResNet-50	ce+dice	CVPR2020	10	0.3111	10	0.3111	0.4114	0.6550
completed	FarSeg	ResNet-50	weighted_ce+dice	CVPR2020	51	0.3073	51	0.3310	0.4607	0.5946
completed	FarSeg	ResNet-50	focal+dice	CVPR2020	5	0.3286	5	0.3286	0.4309	0.7012
completed	BANet	ResT-Lite	ce+dice	RS2021	31	0.2779	31	0.2992	0.3969	0.7172
completed	ABCNet	ResNet-18	ce+dice+aux_ce	ISPRSJPRS2021	12	0.3070	12	0.3070	0.4087	0.7066
completed	MANet	ResNet-50	ce+dice	TGRS2022	31	0.3147	31	0.3147	0.4108	0.6763
completed	MANet	ResNet-50	weighted_ce+dice	TGRS2022	31	0.3079	31	0.3316	0.4447	0.6724
completed	MANet	ResNet-50	focal+dice	TGRS2022	57	0.3015	57	0.3246	0.4414	0.6835
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	ISPRSJPRS2022	5	0.3388	5	0.3388	0.4375	0.7641
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	ISPRSJPRS2022	25	0.3275	25	0.3275	0.4329	0.7243
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	ISPRSJPRS2022	5	0.3314	5	0.3314	0.4291	0.7378
completed	DC-Swin	Swin-Small	ce+dice	TGRS2022	31	0.2884	31	0.2884	0.3846	0.6994
completed	A2FPN	ResNet-18	ce+dice	IJRS2022	18	0.3085	18	0.3085	0.4068	0.6811
completed	A2FPN	ResNet-18	weighted_ce+dice	IJRS2022	12	0.2959	12	0.3187	0.4405	0.6435
completed	A2FPN	ResNet-18	focal+dice	IJRS2022	2	0.3027	2	0.3027	0.4050	0.6693
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	ICASSP2023	80	0.2951	80	0.2951	0.3798	0.7522
completed	FarSeg++	MiT-B2	ce (native)	TGRS2023	65	0.3049	65	0.3284	0.4341	0.7590
completed	SACANet	HRNet-W32	ce+aux_ce (native)	ICME2023	29	0.2985	29	0.3215	0.4191	0.6878
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	GRSL2024	17	0.2772	17	0.2772	0.3665	0.6567
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	GRSL2025	31	0.3362	31	0.3362	0.4369	0.7208
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	GRSL2025	42	0.3466	42	0.3466	0.4708	0.6706
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	GRSL2025	5	0.3411	5	0.3411	0.4402	0.7490
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	GRSL2024	2	0.1559	2	0.1559	0.2179	0.6601
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	GRSL2024	3	0.2313	3	0.2313	0.3219	0.6845
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	GRSL2024	2	0.1910	2	0.1910	0.2521	0.7198
completed	PyramidMamba	Swin-Base	ce+dice	JAG2025	68	0.3703	68	0.3703	0.4900	0.7318
completed	PyramidMamba	Swin-Base	weighted_ce+dice	JAG2025	-	0.3830	-	0.4125	0.5428	0.7538
completed	PyramidMamba	Swin-Base	focal+dice	JAG2025	-	0.3685	-	0.3685	0.4794	0.7263
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	TGRS2025	17	0.2264	17	0.2264	0.3066	0.6353
completed	MF-Mamba	HRNet-W18	ce+dice	TGRS2025	54	0.3039	54	0.3039	0.3988	0.6506
completed	MCPNet	ResNet-50	ce+dice	TGRS2025	60	0.3051	60	0.3051	0.3952	0.7140
completed	MCPNet	ResNet-50	weighted_ce+dice	TGRS2025	23	0.2954	23	0.3181	0.4178	0.7706
completed	MCPNet	ResNet-50	focal+dice	TGRS2025	33	0.3027	33	0.3027	0.3944	0.7601
-	--- METHODS RELATED TO VISION FOUNDATION MODELS ---	-	-	-	-	-	-	-	-	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	NeurIPS2023	74	0.2067	74	0.2226	0.2997	0.6866
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	NeurIPS2023	64	0.2077	64	0.2237	0.3077	0.6653
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	NeurIPS2023	53	0.2087	53	0.2248	0.2973	0.7094
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	TGRS2024	22	0.3104	22	0.3104	0.4073	0.6960
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	TGRS2024	74	0.2909	74	0.2909	0.3898	0.6481
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	TGRS2024	44	0.2859	44	0.2859	0.3777	0.7297
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	TGRS2024	40	0.2971	40	0.2971	0.3934	0.6894
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	ICLR2025	57	0.2082	57	0.2242	0.3053	0.6757
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	ICLR2025	56	0.2160	56	0.2326	0.3165	0.6928
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	ICLR2025	56	0.2124	56	0.2288	0.3108	0.6903
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	ICLR2025	34	0.2903	34	0.2903	0.3880	0.6894
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	ICLR2025	40	0.2723	40	0.2933	0.3976	0.7158
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	ICLR2025	25	0.2906	25	0.2906	0.3862	0.6832
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	RS2025	19	0.2959	19	0.2959	0.3907	0.7304
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	RS2025	13	0.3128	65	0.3333	0.4506	0.6948
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	RS2025	28	0.2841	28	0.2841	0.3742	0.7546
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	TGRS2025	-	0.3094	-	0.3094	0.4078	0.6813
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	TGRS2025	-	0.3035	-	0.3035	0.4059	0.6699
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	TGRS2025	-	0.2984	-	0.3214	0.4445	0.6442
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	TGRS2025	-	0.3078	-	0.3078	0.4096	0.7069
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	TGRS2025	-	0.2791	-	0.3006	0.3987	0.7175
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	TGRS2025	-	0.3162	-	0.3162	0.4125	0.6766
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	TGRS2025	-	0.1723	-	0.3015	0.4057	0.6379
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	TGRS2025	-	0.3085	-	0.3322	0.4453	0.6725
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	TGRS2025	-	0.3399	-	0.3399	0.4388	0.7646
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	TGRS2025	-	0.3406	-	0.3406	0.4403	0.7242
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	TGRS2025	-	0.3286	-	0.3286	0.4342	0.7247

Per-Class IoU Tables (Completed + Running)

Validation (best checkpoint)

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/val_per_class_iou_completed_models_present_only.csv
• Heatmap: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/val_per_class_iou_completed_models_heatmap.png
• Classes shown: 13 (GT-present in val)
• Running rows are placeholders (-) until eval artifacts are generated.

General segmentation models

status	model	backbone	loss	val_miou_present	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.3545	0.2133	0.0864	0.8066	0.2181	0.6615	0.6747	0.0392	0.1743	0.8533	0.3079	0.2300	0.3278	0.0156
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.3750	0.3155	0.1156	0.7247	0.2320	0.6333	0.6592	0.0653	0.4249	0.8360	0.3016	0.2257	0.2937	0.0472
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.3581	0.2864	0.1416	0.7751	0.1851	0.6372	0.6956	0.0379	0.2392	0.8532	0.2776	0.2053	0.3186	0.0026
completed	DeepLabV3+	ResNet-50	ce+dice	0.3379	0.3138	0.0000	0.7313	0.1287	0.5083	0.6617	0.1064	0.3430	0.8274	0.3056	0.2190	0.2469	0.0000
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.3427	0.3405	0.2380	0.7561	0.1523	0.4819	0.6178	0.0660	0.2589	0.8272	0.3118	0.1478	0.2571	0.0000
completed	DeepLabV3+	ResNet-50	focal+dice	0.3461	0.2902	0.1660	0.7753	0.1166	0.5847	0.7103	0.0254	0.1847	0.8476	0.2739	0.1850	0.2795	0.0000
completed	UPerNet	Swin-Tiny	ce+dice	0.2873	0.2280	0.0356	0.6775	0.0000	0.6036	0.3315	0.0475	0.1610	0.8584	0.2953	0.2135	0.2827	0.0000
completed	OCRNet	HRNet-W48	ce+dice	0.2954	0.1587	0.0000	0.7014	0.0000	0.6124	0.3594	0.1218	0.3642	0.7617	0.2700	0.1971	0.2929	0.0000
completed	SegFormer	MiT-B2	ce+dice	0.3480	0.2508	0.0295	0.8176	0.0639	0.5704	0.6613	0.0935	0.3357	0.8436	0.2786	0.2757	0.3028	0.0001
completed	SegFormer	MiT-B2	weighted_ce+dice	0.3423	0.3146	0.0892	0.7779	0.0000	0.6919	0.3261	0.1256	0.4907	0.7972	0.3384	0.1524	0.3460	0.0000
completed	SegFormer	MiT-B2	focal+dice	0.3501	0.2831	0.1476	0.7984	0.0965	0.5890	0.5631	0.0615	0.4210	0.8524	0.2692	0.1864	0.2827	0.0000
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.3033	0.1770	0.1985	0.7404	0.0000	0.4916	0.6668	0.1032	0.1376	0.7004	0.2342	0.2501	0.2438	0.0000
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.2896	0.2133	0.0297	0.7683	0.0000	0.6706	0.2809	0.0468	0.0226	0.8405	0.2988	0.2819	0.2931	0.0000
completed	Afformer	AFFormer-Base	ce+dice	0.3153	0.2372	0.0156	0.7890	0.0000	0.6877	0.6169	0.0559	0.0213	0.8253	0.2894	0.2516	0.3090	0.0000
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.3444	0.2053	0.2414	0.7877	0.0080	0.6405	0.5567	0.0854	0.4163	0.7731	0.2403	0.2299	0.2928	0.0000
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.3840	0.3165	0.2362	0.7856	0.1793	0.5556	0.6910	0.0851	0.3463	0.8189	0.2521	0.1972	0.3088	0.2185
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.3790	0.2128	0.2106	0.7713	0.0638	0.6121	0.6640	0.1234	0.5000	0.8410	0.2837	0.1836	0.2769	0.1845
completed	SeaFormer	SeaFormer-Base	ce+dice	0.3116	0.1769	0.0000	0.7971	0.0000	0.6495	0.6235	0.0463	0.0917	0.8057	0.3248	0.2390	0.2965	0.0000
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.3054	0.2614	0.0312	0.7958	0.0000	0.6478	0.5404	0.0554	0.0285	0.8311	0.2561	0.2336	0.2889	0.0000
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.2549	0.0876	0.1619	0.6831	0.0000	0.4522	0.5112	0.0184	0.0492	0.7612	0.1856	0.1671	0.2366	0.0000

Remote-sensing-specific methods

status	model	backbone	loss	val_miou_present	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice
completed	FarSeg	ResNet-50	ce (native)	0.2989	0.2645	0.0000	0.6644	0.0000	0.6033	0.5275	0.0018	0.3353	0.7271	0.2688	0.2193	0.2743	0.0000
completed	FarSeg	ResNet-50	ce+dice	0.3111	0.3249	0.0000	0.6555	0.0000	0.6419	0.6057	0.0209	0.3463	0.7719	0.1928	0.2226	0.2620	0.0000
completed	FarSeg	ResNet-50	weighted_ce+dice	0.3310	0.2674	0.0716	0.5780	0.1111	0.3713	0.6389	0.1413	0.4714	0.7909	0.1949	0.1558	0.2682	0.2418
completed	FarSeg	ResNet-50	focal+dice	0.3286	0.2404	0.0000	0.7259	0.0000	0.6302	0.6521	0.1548	0.3717	0.8155	0.2369	0.1634	0.2807	0.0000
completed	BANet	ResT-Lite	ce+dice	0.2992	0.1944	0.0000	0.7519	0.0000	0.5869	0.5615	0.1578	0.1030	0.8090	0.2574	0.1872	0.2811	0.0000
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.3070	0.2142	0.0000	0.7417	0.0000	0.5475	0.5567	0.0974	0.2885	0.8266	0.2430	0.1955	0.2807	0.0000
completed	MANet	ResNet-50	ce+dice	0.3147	0.3122	0.0000	0.6774	0.0000	0.6205	0.6587	0.0235	0.3134	0.8350	0.1880	0.1850	0.2778	0.0000
completed	MANet	ResNet-50	weighted_ce+dice	0.3316	0.3611	0.1090	0.6822	0.0910	0.5322	0.6410	0.0338	0.4568	0.7894	0.1837	0.1736	0.2419	0.0139
completed	MANet	ResNet-50	focal+dice	0.3246	0.2882	0.0910	0.6819	0.0264	0.5437	0.6209	0.0143	0.2840	0.7775	0.2534	0.2317	0.2709	0.1343
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.3388	0.3298	0.0000	0.8087	0.0000	0.5542	0.6043	0.0266	0.4635	0.8288	0.3045	0.1869	0.2955	0.0000
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.3275	0.1742	0.1195	0.7758	0.0000	0.5172	0.6793	0.0673	0.4487	0.7556	0.2765	0.1496	0.2931	0.0000
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.3314	0.2782	0.0000	0.7489	0.0000	0.5971	0.7099	0.0291	0.4654	0.7760	0.2662	0.2474	0.2988	0.0000
completed	DC-Swin	Swin-Small	ce+dice	0.2884	0.1573	0.0000	0.7263	0.0000	0.6021	0.4670	0.0950	0.1835	0.8402	0.2047	0.1979	0.2750	0.0000
completed	A2FPN	ResNet-18	ce+dice	0.3085	0.2414	0.0000	0.6983	0.0000	0.5216	0.6605	0.0579	0.3141	0.8409	0.2017	0.1770	0.2970	0.0000
completed	A2FPN	ResNet-18	weighted_ce+dice	0.3187	0.3033	0.0675	0.6520	0.0774	0.3458	0.6664	0.1585	0.3380	0.7892	0.2379	0.1567	0.2730	0.0768
completed	A2FPN	ResNet-18	focal+dice	0.3027	0.3418	0.0000	0.6944	0.0000	0.4617	0.5967	0.0240	0.3705	0.7640	0.2357	0.1630	0.2831	0.0000
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.2951	0.1621	0.0023	0.7084	0.0003	0.2970	0.5871	0.0534	0.0172	0.0416	0.2061	0.1394	0.2864	0.0000
completed	FarSeg++	MiT-B2	ce (native)	0.3284	0.2061	0.1641	0.7977	0.0067	0.6212	0.5303	0.0370	0.3087	0.8464	0.2526	0.2149	0.2834	0.0000
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.3215	0.2771	0.0000	0.6927	0.0000	0.5464	0.6559	0.0007	0.3923	0.8508	0.2915	0.2007	0.2708	0.0000
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.2772	0.2603	0.0000	0.6923	0.0000	0.5261	0.6034	0.0000	0.1737	0.8157	0.1898	0.1209	0.2210	0.0000
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.3362	0.2673	0.0000	0.7545	0.0000	0.5426	0.6552	0.0694	0.4968	0.8348	0.2250	0.2575	0.2673	0.0000
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.3466	0.2916	0.0446	0.6857	0.1125	0.4482	0.6815	0.0605	0.4168	0.7970	0.1750	0.2275	0.2629	0.3021
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.3411	0.2837	0.0000	0.7777	0.0000	0.5865	0.6114	0.0117	0.5306	0.7901	0.3270	0.2119	0.3031	0.0000
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0.1559	0.0000	0.0000	0.7344	0.0000	0.3705	0.0000	0.0025	0.0000	0.4074	0.1691	0.0971	0.2458	0.0000
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.2313	0.0809	0.0000	0.7845	0.0000	0.6007	0.3012	0.0210	0.1857	0.4663	0.1959	0.1665	0.2043	0.0000
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0.1910	0.0000	0.0000	0.8065	0.0000	0.5853	0.0144	0.0136	0.0093	0.5123	0.2080	0.0748	0.2593	0.0000
completed	PyramidMamba	Swin-Base	ce+dice	0.3703	0.3644	0.2072	0.7570	0.1584	0.5820	0.6182	0.0410	0.4457	0.8554	0.2859	0.2096	0.2897	0.0000
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.4125	0.3492	0.2361	0.7829	0.2286	0.6468	0.6794	0.0421	0.5082	0.8528	0.2791	0.2062	0.3292	0.2218
completed	PyramidMamba	Swin-Base	focal+dice	0.3685	0.3503	0.1256	0.7532	0.1062	0.6097	0.7004	0.0331	0.4892	0.8524	0.2937	0.1816	0.2946	0.0000
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.2264	0.0874	0.0000	0.6769	0.0000	0.4898	0.2946	0.0016	0.0581	0.7983	0.1796	0.1273	0.2298	0.0000
completed	MF-Mamba	HRNet-W18	ce+dice	0.3039	0.2858	0.0000	0.6708	0.0000	0.5050	0.6674	0.0052	0.4015	0.8303	0.1578	0.1918	0.2349	0.0000
completed	MCPNet	ResNet-50	ce+dice	0.3051	0.1041	0.0000	0.7422	0.0000	0.5736	0.6592	0.0187	0.3616	0.8390	0.2280	0.2075	0.2332	0.0000
completed	MCPNet	ResNet-50	weighted_ce+dice	0.3181	0.1040	0.0783	0.8021	0.0000	0.6061	0.5622	0.0677	0.2100	0.8293	0.3378	0.2475	0.2906	0.0000
completed	MCPNet	ResNet-50	focal+dice	0.3027	0.1376	0.0000	0.7836	0.0000	0.6150	0.5854	0.0694	0.1373	0.8352	0.2713	0.2263	0.2731	0.0000

Methods related to vision foundation models

status	model	backbone	loss	val_miou_present	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.2226	0.1125	0.0002	0.7274	0.0000	0.5540	0.1486	0.0072	0.0317	0.7444	0.1957	0.1421	0.2298	0.0000
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.2237	0.1148	0.0051	0.7041	0.0363	0.5632	0.1687	0.0383	0.0432	0.7022	0.2138	0.1147	0.2032	0.0000
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.2248	0.1180	0.0046	0.7562	0.0000	0.6064	0.1149	0.0065	0.0128	0.7510	0.1902	0.1303	0.2316	0.0000
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.3104	0.1015	0.0000	0.7134	0.0000	0.6247	0.6719	0.1760	0.2155	0.8072	0.2616	0.1836	0.2759	0.0000
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.2909	0.2431	0.0000	0.6725	0.0000	0.5426	0.6788	0.0305	0.2741	0.7007	0.2317	0.1473	0.2604	0.0000
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.2859	0.1339	0.0000	0.7817	0.0000	0.4701	0.5686	0.0286	0.2696	0.8318	0.1927	0.2125	0.2271	0.0000
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.2971	0.2083	0.0000	0.7157	0.0000	0.4130	0.6491	0.0182	0.3926	0.8202	0.2039	0.2159	0.2254	0.0000
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.2242	0.1131	0.0000	0.7295	0.0000	0.4984	0.2010	0.0047	0.0861	0.7436	0.2142	0.1024	0.2220	0.0000
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.2326	0.0406	0.0308	0.7612	0.0293	0.4463	0.3085	0.0314	0.0990	0.7675	0.2151	0.0861	0.2084	0.0000
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.2288	0.0958	0.0017	0.7499	0.0000	0.4624	0.3107	0.0161	0.0719	0.7424	0.2096	0.1025	0.2108	0.0000
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.2903	0.2215	0.0000	0.7101	0.0000	0.6559	0.3378	0.0272	0.3258	0.8101	0.2173	0.2135	0.2544	0.0000
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.2933	0.1862	0.0943	0.7681	0.0029	0.5044	0.4415	0.0284	0.3490	0.7837	0.2183	0.1966	0.2391	0.0000
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.2906	0.2497	0.0000	0.7281	0.0000	0.5613	0.4376	0.0134	0.3647	0.8286	0.1996	0.1350	0.2599	0.0000
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.2959	0.2226	0.0000	0.7836	0.0000	0.6620	0.3555	0.0263	0.2885	0.8042	0.2741	0.1385	0.2912	0.0000
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.3333	0.2853	0.1837	0.7512	0.0331	0.5979	0.5701	0.0385	0.4021	0.7482	0.2089	0.0936	0.2682	0.1520
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.2841	0.2804	0.0000	0.8023	0.0000	0.6016	0.4233	0.0395	0.1251	0.8108	0.2196	0.1075	0.2833	0.0000
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.3094	0.2442	0.0000	0.6983	0.0000	0.5238	0.6614	0.0597	0.3171	0.8411	0.2016	0.1772	0.2974	0.0000
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.3035	0.3435	0.0000	0.6946	0.0000	0.4638	0.5973	0.0249	0.3706	0.7669	0.2361	0.1646	0.2833	0.0000
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.3214	0.3047	0.0704	0.6523	0.0775	0.3477	0.6681	0.1588	0.3414	0.7920	0.2382	0.1579	0.2731	0.0959
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.3078	0.2153	0.0000	0.7418	0.0000	0.5494	0.5569	0.0984	0.2907	0.8285	0.2431	0.1962	0.2809	0.0000
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.3006	0.2044	0.0000	0.7520	0.0000	0.5884	0.5621	0.1586	0.1050	0.8109	0.2576	0.1881	0.2809	0.0000
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.3162	0.3153	0.0000	0.6773	0.0000	0.6218	0.6591	0.0241	0.3258	0.8354	0.1881	0.1855	0.2779	0.0000
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.3015	0.4533	0.0000	0.7477	0.0559	0.5386	0.0000	0.0142	0.0000	0.0000	0.3457	0.0150	0.2415	0.0000
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.3322	0.3626	0.1104	0.6821	0.0910	0.5331	0.6413	0.0341	0.4602	0.7903	0.1838	0.1738	0.2421	0.0139
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.3399	0.3305	0.0000	0.8089	0.0000	0.5578	0.6052	0.0297	0.4677	0.8301	0.3050	0.1880	0.2959	0.0000
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.3406	0.2782	0.0000	0.7491	0.0000	0.5993	0.7111	0.0310	0.4681	0.7770	0.2665	0.2483	0.2992	0.0000
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.3286	0.1744	0.1238	0.7760	0.0000	0.5211	0.6800	0.0680	0.4502	0.7576	0.2769	0.1504	0.2930	0.0000

Test

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/test_per_class_iou_completed_models_present_only.csv
• Heatmap: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/test_per_class_iou_completed_models_heatmap.png
• Classes shown: 10 (GT-present in test; dropped absent classes: Heavy machinery, Compact mounds, Grass, Vehicles)
• Running rows are placeholders (-) until eval artifacts are generated.

General segmentation models

status	model	backbone	loss	test_miou_present	Building	Mining raft	Primary Forest	Water bodies	Agricultural crop	Gravel mounds	Type1 regen	Type2 regen	Bare ground	Sluice
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.3895	0.3304	0.1906	0.7089	0.7395	0.4176	0.4109	0.2980	0.2636	0.5200	0.0158
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.3766	0.4039	0.1888	0.6503	0.7353	0.2867	0.3817	0.3098	0.2395	0.4885	0.0814
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.3735	0.4205	0.1563	0.6750	0.7615	0.3090	0.3516	0.2940	0.2595	0.5048	0.0025
completed	DeepLabV3+	ResNet-50	ce+dice	0.3662	0.4126	0.0000	0.7100	0.7005	0.3608	0.4825	0.2906	0.2546	0.4506	0.0000
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.3438	0.3376	0.1955	0.6927	0.6909	0.1500	0.3743	0.3336	0.2400	0.4230	0.0000
completed	DeepLabV3+	ResNet-50	focal+dice	0.3515	0.3971	0.1066	0.6884	0.7172	0.2859	0.2962	0.2972	0.2549	0.4715	0.0000
completed	UPerNet	Swin-Tiny	ce+dice	0.3371	0.3163	0.0689	0.6443	0.7172	0.3436	0.3114	0.3008	0.2055	0.4629	0.0000
completed	OCRNet	HRNet-W48	ce+dice	0.2722	0.2982	0.0000	0.5517	0.4619	0.1134	0.3948	0.2974	0.2469	0.3575	0.0000
completed	SegFormer	MiT-B2	ce+dice	0.3222	0.3190	0.1321	0.6619	0.5524	0.2314	0.3474	0.3223	0.2548	0.3886	0.0119
completed	SegFormer	MiT-B2	weighted_ce+dice	0.4010	0.4500	0.1358	0.7137	0.7636	0.3985	0.5275	0.3404	0.1760	0.5043	0.0000
completed	SegFormer	MiT-B2	focal+dice	0.3332	0.3588	0.1843	0.6842	0.4357	0.3276	0.3999	0.3138	0.2634	0.3570	0.0078
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.2985	0.2218	0.1155	0.7008	0.5283	0.2786	0.2635	0.2846	0.2232	0.3682	0.0000
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.2682	0.3313	0.0385	0.6085	0.4889	0.0846	0.2239	0.3168	0.2280	0.3804	0.0000
completed	Afformer	AFFormer-Base	ce+dice	0.3047	0.3622	0.0633	0.6749	0.4837	0.2671	0.2531	0.3398	0.2436	0.3587	0.0004
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.3799	0.3306	0.0949	0.7048	0.7956	0.3462	0.4750	0.3115	0.2275	0.5126	0.0000
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.3790	0.3526	0.1570	0.7041	0.7616	0.3469	0.3744	0.2903	0.2405	0.4831	0.0799
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.3693	0.3134	0.1549	0.6499	0.6826	0.3583	0.4868	0.3036	0.2498	0.4628	0.0311
completed	SeaFormer	SeaFormer-Base	ce+dice	0.3117	0.3221	0.0141	0.6851	0.4595	0.4053	0.3342	0.3298	0.2221	0.3451	0.0000
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.2679	0.3415	0.0863	0.6122	0.4065	0.1962	0.1772	0.3054	0.2107	0.3427	0.0000
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.2789	0.1083	0.0880	0.7338	0.4404	0.3529	0.2254	0.2835	0.2227	0.3336	0.0000

Remote-sensing-specific methods

status	model	backbone	loss	test_miou_present	Building	Mining raft	Primary Forest	Water bodies	Agricultural crop	Gravel mounds	Type1 regen	Type2 regen	Bare ground	Sluice
completed	FarSeg	ResNet-50	ce (native)	0.3564	0.2996	0.0000	0.6518	0.8273	0.2050	0.4567	0.3081	0.2718	0.5442	0.0000
completed	FarSeg	ResNet-50	ce+dice	0.3717	0.3666	0.0000	0.6105	0.8184	0.3664	0.4747	0.3118	0.2522	0.5167	0.0000
completed	FarSeg	ResNet-50	weighted_ce+dice	0.3642	0.3047	0.1328	0.6141	0.7264	0.3078	0.4693	0.2636	0.2445	0.4610	0.1179
completed	FarSeg	ResNet-50	focal+dice	0.3744	0.3855	0.0000	0.6992	0.8519	0.2370	0.4306	0.3358	0.2478	0.5560	0.0000
completed	BANet	ResT-Lite	ce+dice	0.2926	0.2564	0.0000	0.6911	0.5580	0.2844	0.2729	0.2906	0.2283	0.3438	0.0000
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.3145	0.2693	0.0000	0.6888	0.7210	0.1825	0.3820	0.2787	0.1956	0.4275	0.0000
completed	MANet	ResNet-50	ce+dice	0.3711	0.5667	0.0000	0.6558	0.7265	0.3444	0.4506	0.2938	0.2029	0.4700	0.0000
completed	MANet	ResNet-50	weighted_ce+dice	0.3828	0.4332	0.2108	0.6441	0.7291	0.3572	0.4760	0.2907	0.1884	0.4533	0.0440
completed	MANet	ResNet-50	focal+dice	0.3999	0.5479	0.1745	0.6646	0.6925	0.4170	0.4341	0.3045	0.2271	0.4509	0.0848
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.3941	0.4657	0.0000	0.7259	0.7274	0.4635	0.5224	0.3449	0.2181	0.4720	0.0000
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.3566	0.3318	0.1946	0.6817	0.6710	0.2861	0.4238	0.3321	0.2116	0.4327	0.0000
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.3765	0.4652	0.0000	0.7098	0.7641	0.3846	0.4806	0.3143	0.2627	0.4828	0.0000
completed	DC-Swin	Swin-Small	ce+dice	0.2971	0.2815	0.0000	0.6661	0.6159	0.1668	0.3052	0.3172	0.2400	0.3787	0.0000
completed	A2FPN	ResNet-18	ce+dice	0.3688	0.3454	0.0000	0.7085	0.8334	0.2368	0.4817	0.2882	0.2428	0.5512	0.0000
completed	A2FPN	ResNet-18	weighted_ce+dice	0.3720	0.3624	0.1482	0.7198	0.7322	0.3032	0.4070	0.3043	0.2122	0.4455	0.0856
completed	A2FPN	ResNet-18	focal+dice	0.3363	0.3992	0.0000	0.6585	0.6340	0.2202	0.4728	0.3150	0.2345	0.4285	0.0000
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.3108	0.2460	0.1024	0.7280	0.7215	0.2772	0.0170	0.2716	0.2706	0.4524	0.0000
completed	FarSeg++	MiT-B2	ce (native)	0.3062	0.3413	0.1459	0.7048	0.5551	0.1487	0.2526	0.2895	0.2436	0.3800	0.0000
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.3294	0.3359	0.0018	0.6371	0.6208	0.4380	0.2665	0.3319	0.2333	0.4282	0.0000
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.3147	0.2662	0.0731	0.6593	0.6889	0.2644	0.2455	0.2725	0.2254	0.4512	0.0000
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.3520	0.4223	0.0000	0.6743	0.6150	0.3353	0.5308	0.2833	0.2537	0.4057	0.0000
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.3854	0.4859	0.1289	0.6944	0.6466	0.3336	0.4880	0.2873	0.2228	0.4248	0.1412
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.3897	0.4594	0.0000	0.7028	0.7118	0.4937	0.5388	0.3357	0.1800	0.4750	0.0000
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0.2385	0.0000	0.0000	0.7796	0.6756	0.0000	0.0000	0.2615	0.2298	0.4381	0.0000
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.3068	0.1480	0.0000	0.6587	0.5998	0.4101	0.3283	0.2975	0.2396	0.3856	0.0000
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0.2399	0.0000	0.0000	0.7723	0.6926	0.0147	0.0416	0.2654	0.1766	0.4352	0.0000
completed	PyramidMamba	Swin-Base	ce+dice	0.3985	0.6295	0.2154	0.6818	0.6364	0.3555	0.4653	0.3036	0.2522	0.4457	0.0000
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.4414	0.6875	0.2659	0.6891	0.6642	0.4437	0.5345	0.3030	0.2743	0.4459	0.1060
completed	PyramidMamba	Swin-Base	focal+dice	0.3961	0.7059	0.1549	0.6602	0.6184	0.3811	0.4395	0.3284	0.2406	0.4324	0.0000
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.2264	0.1737	0.0165	0.6562	0.5126	0.1640	0.1577	0.2803	0.2214	0.3635	0.0000
completed	MF-Mamba	HRNet-W18	ce+dice	0.3001	0.2849	0.0000	0.6570	0.5295	0.2176	0.4245	0.2855	0.2247	0.3776	0.0000
completed	MCPNet	ResNet-50	ce+dice	0.3056	0.1656	0.0000	0.6890	0.5591	0.2642	0.4480	0.2901	0.2606	0.3795	0.0000
completed	MCPNet	ResNet-50	weighted_ce+dice	0.3193	0.1766	0.1451	0.6789	0.4895	0.3775	0.3701	0.3182	0.2658	0.3710	0.0000
completed	MCPNet	ResNet-50	focal+dice	0.3233	0.1958	0.0000	0.7028	0.6344	0.4144	0.3116	0.3241	0.2410	0.4088	0.0000

Methods related to vision foundation models

status	model	backbone	loss	test_miou_present	Building	Mining raft	Primary Forest	Water bodies	Agricultural crop	Gravel mounds	Type1 regen	Type2 regen	Bare ground	Sluice
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.2485	0.2233	0.0211	0.6664	0.5210	0.0429	0.1300	0.2979	0.2299	0.3530	0.0000
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.2538	0.1961	0.0877	0.6578	0.4813	0.0741	0.1949	0.3270	0.2207	0.2989	0.0000
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.2503	0.2408	0.0521	0.7055	0.5072	0.0366	0.0794	0.2916	0.2327	0.3569	0.0000
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.2964	0.0894	0.0000	0.6257	0.6902	0.2007	0.3997	0.3006	0.2397	0.4174	0.0000
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.2916	0.2048	0.0000	0.6620	0.6011	0.1270	0.3726	0.2936	0.2437	0.4108	0.0000
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.2922	0.2637	0.0000	0.7463	0.5706	0.1557	0.3117	0.2710	0.2614	0.3418	0.0000
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.3241	0.2783	0.0000	0.6849	0.6421	0.1958	0.4495	0.2846	0.2670	0.4384	0.0000
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.2422	0.2552	0.0441	0.6777	0.4286	0.0149	0.2028	0.3060	0.1635	0.3288	0.0000
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.2207	0.0778	0.0403	0.6850	0.3653	0.0532	0.2506	0.2871	0.1351	0.3102	0.0027
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.2351	0.2074	0.0479	0.6960	0.3906	0.0482	0.1869	0.2984	0.1556	0.3204	0.0000
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.2885	0.2944	0.0000	0.6846	0.5456	0.1633	0.2935	0.2949	0.2355	0.3737	0.0000
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.2875	0.1230	0.0774	0.7163	0.6069	0.1829	0.2916	0.2798	0.2168	0.3805	0.0000
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.2980	0.2670	0.0000	0.7298	0.6004	0.1825	0.3277	0.3039	0.1975	0.3712	0.0000
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.3263	0.3254	0.0000	0.7030	0.6784	0.2550	0.3121	0.3022	0.2285	0.4579	0.0000
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.3696	0.3694	0.1755	0.7118	0.6970	0.3144	0.4291	0.3081	0.1970	0.4374	0.0563
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.3450	0.3239	0.0068	0.7604	0.7255	0.2763	0.3453	0.3057	0.2320	0.4741	0.0000
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.3702	0.3508	0.0000	0.7085	0.8338	0.2368	0.4890	0.2883	0.2430	0.5519	0.0000
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.3374	0.4018	0.0000	0.6584	0.6353	0.2203	0.4785	0.3156	0.2347	0.4292	0.0000
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.3745	0.3669	0.1513	0.7198	0.7334	0.3035	0.4118	0.3048	0.2127	0.4463	0.0943
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.3154	0.2720	0.0000	0.6887	0.7224	0.1825	0.3858	0.2794	0.1958	0.4277	0.0000
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.2937	0.2585	0.0000	0.6911	0.5584	0.2844	0.2812	0.2907	0.2286	0.3438	0.0000
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.3604	0.5642	0.0000	0.6667	0.6537	0.2792	0.4624	0.3063	0.1918	0.4800	0.0000
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.4032	0.5654	0.1813	0.6645	0.6930	0.4172	0.4393	0.3048	0.2272	0.4512	0.0883
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.3839	0.4388	0.2110	0.6440	0.7298	0.3573	0.4806	0.2910	0.1887	0.4537	0.0440
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.3958	0.4708	0.0000	0.7259	0.7281	0.4633	0.5327	0.3453	0.2184	0.4735	0.0000
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.3873	0.4662	0.0000	0.7098	0.7649	0.3850	0.4858	0.3149	0.2631	0.4838	0.0000
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.3578	0.3334	0.1955	0.6817	0.6728	0.2858	0.4299	0.3330	0.2123	0.4335	0.0000

Multi-Label Classification from Segmentation

Protocol: image-level labels are derived from predicted and GT segmentation maps (foreground classes 1..14; background 0 ignored).

Validation (best checkpoint)

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/val_multilabel_summary.csv

General segmentation models

status	model	backbone	loss	cp	cr	cf1	op	or	of1	map	macro_f1	micro_f1	sample_f1
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.5172	0.6958	0.5934	0.7024	0.8375	0.764	0.5897	0.5655	0.764	0.8146
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.5329	0.7064	0.6075	0.694	0.8007	0.7436	0.6196	0.5801	0.7436	0.7771
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.5086	0.6714	0.5788	0.6622	0.8605	0.7484	0.5682	0.5517	0.7484	0.7926
completed	DeepLabV3+	ResNet-50	ce+dice	0.4784	0.6032	0.5336	0.6737	0.774	0.7204	0.5264	0.5886	0.7204	0.7493
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.5029	0.6381	0.5625	0.623	0.7884	0.696	0.5553	0.5977	0.696	0.7212
completed	DeepLabV3+	ResNet-50	focal+dice	0.4628	0.6678	0.5468	0.6218	0.8698	0.7252	0.547	0.5706	0.7252	0.7663
completed	UPerNet	Swin-Tiny	ce+dice	0.5578	0.476	0.5137	0.718	0.7352	0.7265	0.4365	0.5761	0.7265	0.7526
completed	OCRNet	HRNet-W48	ce+dice	0.4912	0.4842	0.4877	0.6736	0.7323	0.7017	0.4287	0.6143	0.7017	0.7387
completed	SegFormer	MiT-B2	ce+dice	0.5543	0.6289	0.5892	0.7431	0.791	0.7663	0.5167	0.564	0.7663	0.8121
completed	SegFormer	MiT-B2	weighted_ce+dice	0.5311	0.6078	0.5669	0.735	0.7651	0.7498	0.5279	0.6451	0.7498	0.7977
completed	SegFormer	MiT-B2	focal+dice	0.542	0.6655	0.5974	0.7321	0.8006	0.7649	0.5696	0.5622	0.7649	0.8124
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.5897	0.5054	0.5443	0.6878	0.6533	0.6701	0.4179	0.5605	0.6701	0.683
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.4854	0.4746	0.4799	0.7038	0.7782	0.7391	0.4328	0.5441	0.7391	0.79
completed	Afformer	AFFormer-Base	ce+dice	0.4562	0.5064	0.48	0.7068	0.7778	0.7406	0.4376	0.5147	0.7406	0.7958
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.6174	0.5996	0.6084	0.7255	0.805	0.7632	0.5488	0.6348	0.7632	0.8133
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.5663	0.7099	0.63	0.6949	0.8185	0.7516	0.6285	0.6086	0.7516	0.791
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.5601	0.6726	0.6112	0.6907	0.8029	0.7426	0.5688	0.5772	0.7426	0.7843
completed	SeaFormer	SeaFormer-Base	ce+dice	0.4678	0.4838	0.4756	0.717	0.7808	0.7476	0.4374	0.6066	0.7476	0.7985
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.5395	0.4982	0.518	0.7172	0.7832	0.7487	0.4543	0.5717	0.7487	0.8042
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.6295	0.3932	0.484	0.7052	0.6337	0.6675	0.3697	0.5174	0.6675	0.6841

Remote-sensing-specific methods

status	model	backbone	loss	cp	cr	cf1	op	or	of1	map	macro_f1	micro_f1	sample_f1
completed	FarSeg	ResNet-50	ce (native)	0.4831	0.5141	0.4981	0.6324	0.7799	0.6984	0.4421	0.5917	0.6984	0.7194
completed	FarSeg	ResNet-50	ce+dice	0.4094	0.5747	0.4782	0.5653	0.7885	0.6585	0.4639	0.5753	0.6585	0.6775
completed	FarSeg	ResNet-50	weighted_ce+dice	0.4616	0.7416	0.569	0.5807	0.7627	0.6593	0.5743	0.5203	0.6593	0.6696
completed	FarSeg	ResNet-50	focal+dice	0.4817	0.5744	0.524	0.6525	0.8463	0.7369	0.477	0.6604	0.7369	0.7801
completed	BANet	ResT-Lite	ce+dice	0.4827	0.4964	0.4894	0.7089	0.7754	0.7407	0.4461	0.6205	0.7407	0.7816
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.5471	0.4611	0.5005	0.7536	0.7242	0.7386	0.4279	0.6304	0.7386	0.7741
completed	MANet	ResNet-50	ce+dice	0.3721	0.5759	0.4521	0.5735	0.8223	0.6758	0.4713	0.5728	0.6758	0.705
completed	MANet	ResNet-50	weighted_ce+dice	0.3894	0.7306	0.508	0.5649	0.8231	0.67	0.5744	0.4759	0.67	0.695
completed	MANet	ResNet-50	focal+dice	0.3449	0.7805	0.4784	0.4789	0.8461	0.6116	0.5739	0.4165	0.6116	0.6235
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.454	0.5659	0.5038	0.669	0.868	0.7556	0.4857	0.6341	0.7556	0.7976
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.3739	0.6827	0.4832	0.5765	0.8616	0.6908	0.5253	0.5239	0.6908	0.7424
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.3819	0.5878	0.463	0.5996	0.8484	0.7026	0.4821	0.58	0.7026	0.7387
completed	DC-Swin	Swin-Small	ce+dice	0.4786	0.5122	0.4949	0.6766	0.8113	0.7379	0.4432	0.623	0.7379	0.7789
completed	A2FPN	ResNet-18	ce+dice	0.4653	0.4923	0.4784	0.6713	0.7035	0.687	0.4362	0.5903	0.687	0.7127
completed	A2FPN	ResNet-18	weighted_ce+dice	0.5323	0.6295	0.5768	0.666	0.7097	0.6872	0.4989	0.5572	0.6872	0.7127
completed	A2FPN	ResNet-18	focal+dice	0.3921	0.5559	0.4598	0.5949	0.7816	0.6756	0.4532	0.5535	0.6756	0.7051
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.4844	0.5465	0.5136	0.5995	0.8492	0.7028	0.4002	0.5263	0.7028	0.7472
completed	FarSeg++	MiT-B2	ce (native)	0.6084	0.5488	0.5771	0.7465	0.7864	0.7659	0.4892	0.5457	0.7659	0.812
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.4916	0.4937	0.4927	0.6709	0.755	0.7104	0.4522	0.6073	0.7104	0.7412
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.4681	0.4825	0.4752	0.6278	0.7639	0.6892	0.418	0.5733	0.6892	0.7216
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.4283	0.5533	0.4828	0.6613	0.8225	0.7332	0.4684	0.6124	0.7332	0.7695
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.4391	0.717	0.5447	0.5609	0.8337	0.6706	0.5685	0.5207	0.6706	0.6855
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.4416	0.5687	0.4971	0.6483	0.8582	0.7386	0.4867	0.6014	0.7386	0.7796
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0.1903	0.4219	0.2623	0.4405	0.8515	0.5806	0.3084	0.4605	0.5806	0.6096
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.2515	0.6375	0.3607	0.4168	0.9112	0.572	0.4243	0.4333	0.572	0.5861
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0.3138	0.4805	0.3797	0.5015	0.8785	0.6385	0.3767	0.4873	0.6385	0.6833
completed	PyramidMamba	Swin-Base	ce+dice	0.4913	0.6889	0.5735	0.6552	0.8632	0.7449	0.5754	0.6107	0.7449	0.778
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.4808	0.7874	0.597	0.6368	0.8833	0.7401	0.6522	0.5802	0.7401	0.7892
completed	PyramidMamba	Swin-Base	focal+dice	0.4479	0.6632	0.5347	0.6321	0.8524	0.7259	0.5444	0.527	0.7259	0.7655
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.396	0.454	0.423	0.6091	0.7747	0.682	0.3666	0.5268	0.682	0.7111
completed	MF-Mamba	HRNet-W18	ce+dice	0.457	0.5248	0.4885	0.5991	0.7622	0.6709	0.441	0.5793	0.6709	0.6847
completed	MCPNet	ResNet-50	ce+dice	0.4699	0.4726	0.4712	0.6991	0.7509	0.7241	0.4219	0.5874	0.7241	0.7622
completed	MCPNet	ResNet-50	weighted_ce+dice	0.505	0.5163	0.5106	0.7416	0.793	0.7664	0.4539	0.576	0.7664	0.8162
completed	MCPNet	ResNet-50	focal+dice	0.4861	0.4559	0.4705	0.7492	0.766	0.7575	0.4205	0.593	0.7575	0.808

Methods related to vision foundation models

status	model	backbone	loss	cp	cr	cf1	op	or	of1	map	macro_f1	micro_f1	sample_f1
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.2355	0.5864	0.3361	0.3986	0.9369	0.5592	0.4075	0.3792	0.5592	0.5593
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.2349	0.7073	0.3527	0.3736	0.9328	0.5335	0.4238	0.3532	0.5335	0.5301
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.2626	0.5864	0.3627	0.4156	0.9272	0.5739	0.4167	0.4089	0.5739	0.5847
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.5128	0.4649	0.4877	0.7287	0.7158	0.7222	0.4248	0.6207	0.7222	0.7634
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.5511	0.4662	0.5051	0.6861	0.7295	0.7071	0.4231	0.6135	0.7071	0.7282
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.5192	0.4578	0.4866	0.7222	0.7416	0.7317	0.42	0.5871	0.7317	0.7774
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.4133	0.519	0.4602	0.6086	0.7911	0.6879	0.4381	0.5727	0.6879	0.7149
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.2597	0.5792	0.3586	0.4626	0.9134	0.6142	0.4036	0.4507	0.6142	0.6247
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.2663	0.7744	0.3963	0.4428	0.9136	0.5965	0.4245	0.3837	0.5965	0.6216
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.3021	0.6078	0.4036	0.485	0.9138	0.6336	0.419	0.4576	0.6336	0.6419
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.4659	0.4935	0.4793	0.7006	0.749	0.724	0.4392	0.5926	0.724	0.7614
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.455	0.6322	0.5292	0.6422	0.8003	0.7126	0.4711	0.4992	0.7126	0.7651
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.4567	0.508	0.481	0.6559	0.7778	0.7117	0.4388	0.5864	0.7117	0.7417
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.3897	0.5442	0.4542	0.6344	0.825	0.7173	0.4468	0.561	0.7173	0.7767
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.4677	0.7451	0.5747	0.6376	0.8058	0.7119	0.5729	0.5353	0.7119	0.7611
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.4602	0.5203	0.4884	0.6902	0.8206	0.7498	0.4528	0.605	0.7498	0.8055
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.4653	0.4923	0.4784	0.6713	0.7035	0.687	0.4363	0.5903	0.687	0.7127
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.3921	0.5559	0.4598	0.5948	0.7817	0.6755	0.4528	0.5535	0.6755	0.705
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.5323	0.6295	0.5768	0.666	0.7097	0.6872	0.4991	0.5572	0.6872	0.7127
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.5471	0.4611	0.5005	0.7536	0.7242	0.7386	0.4279	0.6304	0.7386	0.7741
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.4827	0.4964	0.4894	0.7089	0.7754	0.7407	0.4463	0.6205	0.7407	0.7816
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.3721	0.5759	0.4521	0.5735	0.8223	0.6758	0.4717	0.5728	0.6758	0.705
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.5954	0.7652	0.6697	0.525	0.8421	0.6467	0.7194	0.6368	0.6467	0.6806
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.3894	0.7306	0.508	0.5649	0.8231	0.67	0.5747	0.4759	0.67	0.695
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.454	0.5659	0.5038	0.669	0.868	0.7556	0.4858	0.6341	0.7556	0.7976
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.3819	0.5878	0.463	0.5996	0.8484	0.7026	0.4823	0.58	0.7026	0.7387
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.3739	0.6827	0.4832	0.5765	0.8616	0.6908	0.526	0.5239	0.6908	0.7424

Validation per-class precision

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/val_multilabel_per_class_precision.csv

General segmentation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.4202	0.2679	0.866	0.1364	0.6004	0.6751	0.3862	0.4334	0.8869	0.5776	0.6026	0.574	0.2973	0
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.3951	0.2821	0.8872	0.1622	0.6671	0.7566	0.5163	0.3993	0.8195	0.5646	0.5529	0.592	0.3333	0
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.4234	0.1846	0.8525	0.1538	0.5412	0.8186	0.3585	0.49	0.862	0.5021	0.5393	0.5785	0.3077	0
completed	DeepLabV3+	ResNet-50	ce+dice	0.4556	0	0.8427	0.0882	0.5997	0.8339	0.6548	0.1756	0.8323	0.6415	0.549	0.5454	0	0
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.4754	0.5909	0.8653	0.5455	0.5041	0.6627	0.3323	0.2994	0.8145	0.4766	0.582	0.3895	0	0
completed	DeepLabV3+	ResNet-50	focal+dice	0.2705	0.2034	0.8481	0.5	0.5069	0.5683	0.5	0.374	0.7628	0.517	0.4694	0.4964	0	0
completed	UPerNet	Swin-Tiny	ce+dice	0.5846	0.375	0.8861	0	0.678	0.803	0.6947	0.531	0.9603	0.586	0.5735	0.579	0	0
completed	OCRNet	HRNet-W48	ce+dice	0.5	0	0.9178	0	0.6756	0.7274	0.5918	0.487	0.95	0.5317	0.5019	0.5028	0	0
completed	SegFormer	MiT-B2	ce+dice	0.4113	0.1379	0.9128	0.0612	0.7038	0.6473	0.7063	0.5282	0.947	0.5428	0.6243	0.6614	0.3214	0
completed	SegFormer	MiT-B2	weighted_ce+dice	0.4064	0.4848	0.8774	0	0.723	0.8926	0.5206	0.3499	0.8348	0.5973	0.573	0.6448	0	0
completed	SegFormer	MiT-B2	focal+dice	0.187	0.4	0.8804	0.2069	0.7493	0.7559	0.6776	0.5046	0.7848	0.6341	0.5745	0.5836	0.1071	0
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.6061	0.3846	0.9306	0.3333	0.6304	0.9759	0.6162	0.6333	0.8646	0.4536	0.53	0.7075	0	0
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.3778	0.375	0.9036	0	0.6577	0.7125	0.4306	0.3333	0.8542	0.5726	0.5522	0.541	0	0
completed	Afformer	AFFormer-Base	ce+dice	0.4035	0.2143	0.8932	0.125	0.6865	0.6222	0.2361	0.1854	0.8488	0.6047	0.5392	0.5719	0	0
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.5543	1	0.8738	0.6	0.7022	0.6908	0.6848	0.3818	0.7555	0.6123	0.5785	0.5923	0	0
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.3616	0.2826	0.8809	0.2727	0.5278	0.716	0.6	0.5145	0.8603	0.5658	0.5726	0.6166	0.5909	0
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.4787	0.2979	0.8895	0.102	0.6511	0.7831	0.6882	0.3078	0.9233	0.5338	0.5478	0.5152	0.5625	0
completed	SeaFormer	SeaFormer-Base	ce+dice	0.375	0	0.8884	0	0.6919	0.8463	0.2657	0.4489	0.8694	0.5811	0.5915	0.5233	0	0
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.3261	1	0.8878	0	0.722	0.597	0.3485	0.614	0.8117	0.5897	0.5754	0.5408	0	0
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.5714	0.8333	0.9412	0	0.7244	0.9624	0.8684	0.6575	0.9515	0.427	0.5492	0.697	0	0

Remote-sensing-specific methods

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	FarSeg	ResNet-50	ce (native)	0.5385	0	0.8403	0	0.536	0.7215	0.75	0.3577	0.9978	0.4944	0.544	0.4997	0	0
completed	FarSeg	ResNet-50	ce+dice	0.4843	0	0.8875	0	0.6307	0.4296	0.5663	0.2604	0.8109	0.3836	0.5228	0.3468	0	0
completed	FarSeg	ResNet-50	weighted_ce+dice	0.2642	0.1532	0.9158	0.1053	0.4659	0.506	0.6324	0.2581	0.7374	0.3551	0.5398	0.5849	0.4828	0
completed	FarSeg	ResNet-50	focal+dice	0.3939	0	0.8653	0	0.5983	0.8312	0.607	0.5774	0.8789	0.4344	0.5661	0.51	0	0
completed	BANet	ResT-Lite	ce+dice	0.3789	0	0.8832	0	0.694	0.8204	0.6583	0.3478	0.8021	0.5506	0.5713	0.5682	0	0
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.5147	0	0.9084	0	0.7447	0.8941	0.7111	0.597	0.9199	0.5982	0.5711	0.6532	0	0
completed	MANet	ResNet-50	ce+dice	0.3146	0	0.8238	0	0.4487	0.6193	0.3878	0.3004	0.5268	0.4755	0.4691	0.4721	0	0
completed	MANet	ResNet-50	weighted_ce+dice	0.4731	0.0963	0.8188	0.037	0.3772	0.5354	0.3649	0.2771	0.4867	0.4954	0.5144	0.5133	0.0729	0
completed	MANet	ResNet-50	focal+dice	0.3364	0.0411	0.8349	0.0353	0.3977	0.5648	0.5538	0.1614	0.198	0.4294	0.4642	0.3309	0.1359	0
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.4114	0	0.8466	0	0.521	0.9171	0.3784	0.3625	0.7982	0.5515	0.6135	0.5023	0	0
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.1257	0.2278	0.8474	0	0.5212	0.744	0.4138	0.1757	0.2598	0.4653	0.5556	0.5239	0	0
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.3671	0	0.8942	0	0.5375	0.8089	0.2328	0.3293	0.2631	0.5029	0.5641	0.4653	0	0
completed	DC-Swin	Swin-Small	ce+dice	0.58	0	0.8759	0	0.6234	0.7444	0.4359	0.4116	0.9603	0.5302	0.5349	0.5257	0	0
completed	A2FPN	ResNet-18	ce+dice	0.4044	0	0.9053	0	0.5245	0.8696	0.568	0.1716	0.9016	0.5726	0.5307	0.6011	0	0
completed	A2FPN	ResNet-18	weighted_ce+dice	0.4602	0.3846	0.9233	0.1724	0.4866	0.7904	0.5919	0.3402	0.8715	0.5404	0.5273	0.599	0.2326	0
completed	A2FPN	ResNet-18	focal+dice	0.4823	0	0.909	0	0.5174	0.5923	0.5138	0.1813	0.3968	0.4271	0.596	0.4814	0	0
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.4455	0.1429	0.8224	0.75	0.4358	0.7314	0.3972	0.4651	0.6817	0.442	0.5063	0.4765	0	0
completed	FarSeg++	MiT-B2	ce (native)	0.4466	0.3333	0.8748	0.2	0.6713	0.8393	0.6222	0.6627	0.9153	0.6526	0.6103	0.5807	0.5	0
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.6875	0	0.8748	0	0.673	0.8712	0.4474	0.5636	0.7381	0.5333	0.5278	0.4746	0	0
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.5301	0.5	0.8429	0	0.5178	0.6987	0	0.784	0.7433	0.4792	0.5348	0.4544	0	0
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.3905	0	0.8969	0	0.5439	0.6673	0.4118	0.3706	0.657	0.5494	0.5197	0.5606	0	0
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.3319	0.2059	0.8872	0.15	0.422	0.4974	0.51	0.3189	0.4114	0.3551	0.5152	0.4949	0.6087	0
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.4153	0	0.8454	0	0.4969	0.9383	0.5294	0.3434	0.4915	0.5017	0.6318	0.5467	0	0
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0	0	0.8251	0	0.3087	0	0.097	0	0.1374	0.4281	0.3733	0.3043	0	0
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.0647	0	0.8464	0	0.5067	0.3477	0.2275	0.0476	0.1831	0.3494	0.4931	0.2036	0	0
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0	0	0.8185	0	0.4912	0.8426	0.213	0.312	0.1446	0.4323	0.4778	0.3479	0	0
completed	PyramidMamba	Swin-Base	ce+dice	0.4379	0.3438	0.8505	0.2609	0.5838	0.6875	0.3647	0.4737	0.7852	0.5366	0.5146	0.5472	0	0
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.407	0.2769	0.8355	0.1875	0.5518	0.4582	0.2981	0.2816	0.7615	0.5215	0.5373	0.5831	0.55	0
completed	PyramidMamba	Swin-Base	focal+dice	0.2664	0.0439	0.8361	0.0556	0.6371	0.7443	0.6316	0.3033	0.8337	0.5171	0.531	0.4223	0	0
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.3191	0	0.881	0	0.5515	0.5512	0.1864	0.4881	0.7898	0.3757	0.5161	0.4888	0	0
completed	MF-Mamba	HRNet-W18	ce+dice	0.5243	0	0.8887	0	0.6276	0.6373	0.7931	0.2814	0.8495	0.3686	0.5205	0.4494	0	0
completed	MCPNet	ResNet-50	ce+dice	0.2564	0	0.8782	0	0.6446	0.8547	0.4043	0.5247	0.9191	0.5912	0.5641	0.4719	0	0
completed	MCPNet	ResNet-50	weighted_ce+dice	0.4286	0.087	0.8816	0	0.6672	0.8412	0.4348	0.4656	0.8938	0.6224	0.6488	0.5942	0	0
completed	MCPNet	ResNet-50	focal+dice	0.151	0	0.8767	0	0.685	0.9109	0.463	0.4015	0.9695	0.6431	0.6303	0.5878	0	0

Methods related to vision foundation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.1538	0.0909	0.8171	0	0.3927	0.1226	0.1424	0.1163	0.2996	0.2572	0.3201	0.3488	0	0
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.1293	0.0479	0.8373	0.0165	0.372	0.0944	0.2645	0.121	0.2228	0.2512	0.3251	0.3718	0	0
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.3438	0.12	0.83	0	0.3798	0.1432	0.1861	0.1507	0.3156	0.2696	0.3343	0.3402	0	0
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.4865	0	0.9335	0	0.7403	0.9701	0.5975	0.2775	0.9854	0.5209	0.5806	0.5744	0	0
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.5882	0	0.9004	0	0.6748	0.9533	0.7857	0.7578	0.9568	0.4869	0.5042	0.5562	0	0
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.5122	0	0.8997	0	0.5544	0.7933	0.8824	0.4653	0.8673	0.6044	0.6062	0.5645	0	0
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.4464	0	0.8341	0	0.4672	0.5884	0.382	0.5993	0.5725	0.4643	0.5651	0.4539	0	0
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.1855	0	0.8074	0	0.4709	0.2084	0.1457	0.144	0.3253	0.2782	0.4139	0.3969	0	0
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.0533	0.0337	0.8235	0.0141	0.5108	0.1335	0.2759	0.1092	0.317	0.3233	0.4558	0.412	0	0
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.1197	0.1364	0.8121	0	0.5321	0.194	0.2849	0.1881	0.552	0.2884	0.426	0.3941	0	0
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.3945	0	0.8804	0	0.7228	0.7519	0.4815	0.2929	0.8296	0.5785	0.5538	0.5708	0	0
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.3609	0.2078	0.8565	0.0077	0.552	0.5118	0.8571	0.2168	0.687	0.5066	0.624	0.5274	0	0
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.3392	0	0.8877	0	0.6847	0.5629	0.5897	0.4054	0.9165	0.3673	0.6132	0.5702	0	0
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.2336	0	0.8866	0	0.6964	0.4625	0.4417	0.2015	0.6843	0.4617	0.5249	0.4731	0	0
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.4286	0.2051	0.8933	0.0674	0.626	0.8118	0.3783	0.3059	0.656	0.4563	0.5283	0.4631	0.2593	0
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.3439	0	0.84	0	0.6892	0.7429	0.5618	0.3434	0.808	0.5104	0.6251	0.5175	0	0
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.4044	0	0.9053	0	0.5245	0.8696	0.568	0.1716	0.9016	0.5726	0.5307	0.6011	0	0
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.4823	0	0.909	0	0.5169	0.5923	0.5138	0.1813	0.3968	0.4271	0.596	0.4814	0	0
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.4602	0.3846	0.9233	0.1724	0.4865	0.7904	0.5919	0.3402	0.8715	0.5404	0.5273	0.599	0.2326	0
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.5147	0	0.9084	0	0.7443	0.8941	0.7111	0.597	0.9199	0.5982	0.5711	0.6532	0	0
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.3789	0	0.8831	0	0.694	0.8204	0.6583	0.3478	0.8021	0.5506	0.5713	0.5682	0	0
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.3146	0	0.8238	0	0.4487	0.6193	0.3878	0.3004	0.5268	0.4755	0.4691	0.4721	0	0
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.963	0	0.6205	0.2857	0.4143	0	0.5618	0	0	0.9186	0.5112	0.488	0	0
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.4731	0.0963	0.8188	0.037	0.3772	0.5354	0.3649	0.2771	0.4867	0.4954	0.5144	0.5133	0.0729	0
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.4114	0	0.8466	0	0.521	0.9171	0.3784	0.3625	0.7982	0.5515	0.6135	0.5022	0	0
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.3671	0	0.8942	0	0.5373	0.8089	0.2328	0.3293	0.2631	0.5029	0.5641	0.4654	0	0
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.1257	0.2278	0.8474	0	0.5212	0.744	0.4138	0.1757	0.2598	0.4653	0.5556	0.5239	0	0

Validation per-class recall

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/val_multilabel_per_class_recall.csv

General segmentation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.4854	0.8333	0.9363	0.6667	0.9067	0.7757	0.1077	0.4676	0.9741	0.7605	0.6851	0.8352	0.6111	-
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.6214	0.6111	0.809	0.6667	0.8631	0.7193	0.1519	0.6873	0.9777	0.8111	0.7759	0.8215	0.6667	-
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.5631	0.6667	0.9405	0.6667	0.9098	0.7173	0.1462	0.4817	0.9714	0.8194	0.7741	0.8493	0.2222	-
completed	DeepLabV3+	ResNet-50	ce+dice	0.7476	0	0.8286	0.6667	0.897	0.7123	0.2481	0.6676	0.9759	0.6737	0.6721	0.7519	0	-
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.5631	0.7222	0.8856	0.6667	0.8448	0.7314	0.2115	0.5634	0.9768	0.8291	0.4992	0.8018	0	-
completed	DeepLabV3+	ResNet-50	focal+dice	0.6408	0.6667	0.9501	0.5556	0.9076	0.7948	0.1827	0.5521	0.983	0.8224	0.779	0.847	0	-
completed	UPerNet	Swin-Tiny	ce+dice	0.3689	0.1667	0.7645	0	0.8423	0.4266	0.1269	0.338	0.9517	0.7675	0.658	0.7766	0	-
completed	OCRNet	HRNet-W48	ce+dice	0.2427	0	0.7308	0	0.782	0.4457	0.4154	0.4761	0.8651	0.7686	0.6918	0.8767	0	-
completed	SegFormer	MiT-B2	ce+dice	0.5631	0.4444	0.8874	0.3333	0.8121	0.7827	0.2173	0.5803	0.958	0.688	0.6954	0.7135	0.5	-
completed	SegFormer	MiT-B2	weighted_ce+dice	0.7379	0.8889	0.8893	0	0.8597	0.4517	0.4135	0.7549	0.9392	0.8084	0.3941	0.7633	0	-
completed	SegFormer	MiT-B2	focal+dice	0.835	0.6667	0.9386	0.6667	0.8162	0.7072	0.1981	0.6169	0.9777	0.617	0.6039	0.8409	0.1667	-
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.3883	0.5556	0.6252	0.3333	0.7369	0.6117	0.1173	0.2676	0.9187	0.6238	0.743	0.6488	0	-
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.3301	0.1667	0.844	0	0.8398	0.5262	0.1192	0.1268	0.9526	0.685	0.7728	0.8067	0	-
completed	Afformer	AFFormer-Base	ce+dice	0.4466	0.1667	0.8672	0.2222	0.8317	0.6529	0.2038	0.0789	0.958	0.6571	0.7028	0.7957	0	-
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.4951	0.5556	0.9308	0.3333	0.856	0.7575	0.2173	0.6141	0.9777	0.5988	0.6641	0.7945	0	-
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.6214	0.7222	0.9159	0.6667	0.8896	0.7686	0.2192	0.5493	0.9741	0.7218	0.6652	0.7919	0.7222	-
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.4369	0.7778	0.8933	0.5556	0.8404	0.7193	0.225	0.6225	0.9678	0.7608	0.6193	0.8246	0.5	-
completed	SeaFormer	SeaFormer-Base	ce+dice	0.4369	0	0.8745	0	0.8215	0.659	0.1462	0.2225	0.9455	0.6942	0.6558	0.833	0	-
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.4369	0.3333	0.8902	0	0.8128	0.6378	0.1327	0.0986	0.9705	0.6737	0.6658	0.8238	0	-
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.1553	0.2778	0.6723	0	0.6501	0.4125	0.0635	0.1352	0.8758	0.6883	0.6301	0.5506	0	-

Remote-sensing-specific methods

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	FarSeg	ResNet-50	ce (native)	0.4757	0	0.76	0	0.8977	0.6439	0.0115	0.6056	0.8034	0.7772	0.8468	0.8615	0	-
completed	FarSeg	ResNet-50	ce+dice	0.7476	0	0.7228	0	0.8862	0.7394	0.0904	0.7606	0.9312	0.8406	0.8631	0.8896	0	-
completed	FarSeg	ResNet-50	weighted_ce+dice	0.6796	0.9444	0.7342	0.6667	0.8986	0.7163	0.225	0.7211	0.9714	0.8506	0.6992	0.7561	0.7778	-
completed	FarSeg	ResNet-50	focal+dice	0.7573	0	0.906	0	0.9023	0.659	0.3654	0.5042	0.9142	0.8965	0.7213	0.8413	0	-
completed	BANet	ResT-Lite	ce+dice	0.3495	0	0.851	0	0.8345	0.5976	0.2519	0.4507	0.9634	0.7092	0.6887	0.7561	0	-
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.3398	0	0.7923	0	0.7474	0.5946	0.1846	0.338	0.9446	0.6611	0.6395	0.7527	0	-
completed	MANet	ResNet-50	ce+dice	0.6505	0	0.8318	0	0.9216	0.7414	0.2558	0.6761	0.9678	0.7487	0.8381	0.8554	0	-
completed	MANet	ResNet-50	weighted_ce+dice	0.767	1	0.8516	0.4444	0.917	0.7978	0.2962	0.6986	0.9777	0.7683	0.7609	0.8291	0.3889	-
completed	MANet	ResNet-50	focal+dice	0.699	1	0.8473	0.6667	0.9222	0.7757	0.1981	0.7042	0.9848	0.8538	0.8258	0.8912	0.7778	-
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.699	0	0.964	0	0.9095	0.668	0.1346	0.5718	0.9508	0.8181	0.7459	0.895	0	-
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.8835	1	0.9237	0	0.8815	0.7455	0.2308	0.738	0.9812	0.8852	0.7352	0.8706	0	-
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.7379	0	0.8616	0	0.917	0.7284	0.2212	0.6197	0.9812	0.8218	0.8461	0.9068	0	-
completed	DC-Swin	Swin-Small	ce+dice	0.2816	0	0.8638	0	0.8837	0.5332	0.2288	0.5577	0.9508	0.7267	0.8204	0.8124	0	-
completed	A2FPN	ResNet-18	ce+dice	0.534	0	0.7313	0	0.8743	0.671	0.1365	0.5549	0.958	0.5533	0.6301	0.7565	0	-
completed	A2FPN	ResNet-18	weighted_ce+dice	0.5049	0.2778	0.7106	0.5556	0.8106	0.7284	0.4212	0.6028	0.9517	0.6963	0.5918	0.7759	0.5556	-
completed	A2FPN	ResNet-18	focal+dice	0.6602	0	0.8001	0	0.8411	0.7163	0.1077	0.7606	0.9821	0.8594	0.6236	0.8752	0	-
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.4369	0.2222	0.9728	0.3333	0.9039	0.7123	0.2712	0.169	0.714	0.8379	0.6833	0.8478	0	-
completed	FarSeg++	MiT-B2	ce (native)	0.4466	0.4444	0.9235	0.2222	0.8423	0.6358	0.1077	0.4704	0.9651	0.5937	0.6316	0.7953	0.0556	-
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.4272	0	0.7868	0	0.7689	0.7012	0.0327	0.4366	0.9723	0.7788	0.6643	0.8493	0	-
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.4272	0.0556	0.8106	0	0.856	0.672	0	0.2761	0.9473	0.7466	0.6779	0.8037	0	-
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.6408	0	0.8706	0	0.8532	0.7485	0.1481	0.569	0.9705	0.7154	0.839	0.8379	0	-
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.7476	0.3889	0.8817	0.6667	0.8563	0.7716	0.2442	0.6028	0.9857	0.7949	0.8085	0.7938	0.7778	-
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.7379	0	0.9562	0	0.9079	0.6579	0.0519	0.7042	0.9768	0.8568	0.6898	0.8535	0	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0	0	0.9743	0	0.8383	0	0.0942	0	0.9723	0.8594	0.8112	0.9349	0	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.8641	0	0.9481	0	0.9054	0.7847	0.5404	0.5296	0.9517	0.9605	0.8495	0.954	0	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0	0	0.9797	0	0.8554	0.3662	0.3096	0.1099	0.9732	0.8971	0.8081	0.9475	0	-
completed	PyramidMamba	Swin-Base	ce+dice	0.7184	0.6111	0.9039	0.6667	0.9002	0.7636	0.2308	0.6592	0.9705	0.8288	0.8571	0.8455	0	-
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.7864	1	0.9502	0.6667	0.9201	0.8048	0.2981	0.7352	0.9732	0.8457	0.8128	0.8314	0.6111	-
completed	PyramidMamba	Swin-Base	focal+dice	0.7864	0.2778	0.9021	0.6667	0.8616	0.7555	0.1385	0.7296	0.9589	0.8603	0.7502	0.9346	0	-
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.2913	0	0.8391	0	0.8121	0.3954	0.0635	0.231	0.9401	0.8164	0.7036	0.8101	0	-
completed	MF-Mamba	HRNet-W18	ce+dice	0.5243	0	0.7366	0	0.7972	0.7052	0.0442	0.6056	0.9634	0.821	0.7828	0.8417	0	-
completed	MCPNet	ResNet-50	ce+dice	0.3883	0	0.8476	0	0.81	0.6509	0.0365	0.3887	0.9643	0.6439	0.5813	0.8318	0	-
completed	MCPNet	ResNet-50	weighted_ce+dice	0.2621	0.5556	0.9156	0	0.8243	0.5704	0.1731	0.3437	0.9473	0.7213	0.6665	0.7321	0	-
completed	MCPNet	ResNet-50	focal+dice	0.3592	0	0.9147	0	0.823	0.5654	0.1442	0.3042	0.882	0.5472	0.6701	0.7161	0	-

Methods related to vision foundation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.4078	0.1111	0.9792	0	0.9518	0.8068	0.1865	0.3493	0.9651	0.9747	0.9582	0.9323	0.0000	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.3981	0.4444	0.9697	0.6667	0.9369	0.8893	0.5346	0.6028	0.9669	0.9796	0.8944	0.9117	0.0000	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.4272	0.1667	0.9737	0	0.9353	0.7907	0.3558	0.1859	0.9660	0.9532	0.9336	0.9357	0.0000	-
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.3711	0.0000	0.768	0	0.7661	0.6331	0.2712	0.2845	0.9099	0.792	0.5597	0.6878	0.0000	-
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.3883	0.0000	0.7663	0	0.7913	0.6368	0.0846	0.3437	0.8901	0.7713	0.6388	0.75	0.0000	-
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.2039	0.0000	0.8704	0	0.8367	0.6408	0.0577	0.5662	0.9464	0.5337	0.611	0.6842	0.0000	-
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.4854	0.0000	0.8433	0	0.8731	0.7666	0.0654	0.4845	0.9705	0.7081	0.7177	0.8326	0.0000	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.4951	0.0000	0.987	0	0.9045	0.6539	0.1538	0.6507	0.9589	0.9557	0.8569	0.9136	0.0000	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.8155	0.8333	0.9813	0.8889	0.8523	0.8129	0.4558	0.7690	0.9768	0.9645	0.8126	0.9045	0.0000	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.6311	0.1667	0.9862	0	0.8967	0.7696	0.2788	0.5099	0.9535	0.9643	0.8267	0.9174	0.0000	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.4175	0.0000	0.7977	0	0.8093	0.4970	0.075	0.7211	0.9267	0.5982	0.7777	0.7957	0.0000	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.466	0.8889	0.925	0.5556	0.8149	0.6529	0.0692	0.7690	0.9571	0.7334	0.609	0.777	0.0000	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.5631	0.0000	0.8887	0	0.7673	0.5986	0.0442	0.6338	0.9508	0.8495	0.5238	0.7839	0.0000	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.5534	0.0000	0.9093	0	0.842	0.5453	0.1385	0.7634	0.9374	0.8326	0.6804	0.8729	0.0000	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.6699	0.8889	0.8701	0.6667	0.8778	0.6640	0.2212	0.7972	0.9750	0.8621	0.5659	0.8505	0.7778	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.6311	0.0000	0.9675	0	0.847	0.6076	0.0962	0.4789	0.9330	0.7923	0.531	0.8794	0.0000	-
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.534	0.0000	0.7313	0	0.8743	0.6710	0.1365	0.5549	0.9580	0.5533	0.6301	0.7565	0.0000	-
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.6602	0.0000	0.8001	0	0.8414	0.7163	0.1077	0.7606	0.9821	0.8594	0.6236	0.8752	0.0000	-
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.5049	0.2778	0.7106	0.5556	0.8106	0.7284	0.4212	0.6028	0.9517	0.6963	0.5918	0.7759	0.5556	-
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.3398	0.0000	0.7923	0	0.7477	0.5946	0.1846	0.3380	0.9446	0.6611	0.6395	0.7527	0.0000	-
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.3495	0.0000	0.851	0	0.8345	0.5976	0.2519	0.4507	0.9634	0.7092	0.6887	0.7561	0.0000	-
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.6505	0.0000	0.8318	0	0.9216	0.7414	0.2558	0.6761	0.9678	0.7487	0.8381	0.8554	0.0000	-
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.8387	-	1	0.6667	0.9363	-	0.3106	-	-	0.9471	0.4405	0.9819	-	-
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.767	1.0000	0.8516	0.4444	0.917	0.7978	0.2962	0.6986	0.9777	0.7683	0.7609	0.8291	0.3889	-
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.699	0.0000	0.964	0	0.9095	0.6680	0.1346	0.5718	0.9508	0.8181	0.7459	0.895	0.0000	-
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.7379	0.0000	0.8616	0	0.917	0.7284	0.2212	0.6197	0.9812	0.8218	0.8461	0.9072	0.0000	-
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.8835	1.0000	0.9237	0	0.8818	0.7455	0.2308	0.7380	0.9812	0.8852	0.7352	0.8706	0.0000	-

Validation per-class f1

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/val_multilabel_per_class_f1.csv

General segmentation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.4505	0.4054	0.8998	0.2264	0.7224	0.7219	0.1684	0.4499	0.9284	0.6565	0.6412	0.6804	0.4000	-
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.483	0.3860	0.8463	0.2609	0.7525	0.7375	0.2348	0.5052	0.8916	0.6658	0.6457	0.6881	0.4444	-
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.4833	0.2892	0.8943	0.2500	0.6787	0.7646	0.2077	0.4858	0.9134	0.6227	0.6357	0.6883	0.2581	-
completed	DeepLabV3+	ResNet-50	ce+dice	0.5662	-	0.8356	0.1558	0.7188	0.7683	0.3598	0.278	0.8984	0.6572	0.6043	0.6322	-	-
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.5156	0.6500	0.8753	0.6000	0.6314	0.6954	0.2585	0.391	0.8883	0.6053	0.5374	0.5243	-	-
completed	DeepLabV3+	ResNet-50	focal+dice	0.3804	0.3117	0.8962	0.5263	0.6505	0.6628	0.2676	0.446	0.859	0.6349	0.5858	0.626	-	-
completed	UPerNet	Swin-Tiny	ce+dice	0.4524	0.2308	0.8208	-	0.7513	0.5572	0.2146	0.4131	0.956	0.6646	0.6129	0.6634	-	-
completed	OCRNet	HRNet-W48	ce+dice	0.3268	-	0.8137	-	0.7249	0.5527	0.4881	0.4815	0.9055	0.6286	0.5818	0.6391	-	-
completed	SegFormer	MiT-B2	ce+dice	0.4754	0.2105	0.8999	0.1034	0.7541	0.7086	0.3324	0.553	0.9525	0.6069	0.6579	0.6864	0.3913	-
completed	SegFormer	MiT-B2	weighted_ce+dice	0.5241	0.6275	0.8833	-	0.7855	0.5999	0.4609	0.4781	0.8839	0.687	0.467	0.6991	-	-
completed	SegFormer	MiT-B2	focal+dice	0.3055	0.5000	0.9085	0.3158	0.7813	0.7308	0.3065	0.5551	0.8707	0.6254	0.5888	0.689	0.1304	-
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.4734	0.4545	0.7479	0.3333	0.6795	0.752	0.1971	0.3762	0.8908	0.5252	0.6187	0.6769	-	-
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.3523	0.2308	0.8728	-	0.7377	0.6053	0.1867	0.1837	0.9007	0.6238	0.6441	0.6476	-	-
completed	Afformer	AFFormer-Base	ce+dice	0.424	0.1875	0.88	0.1600	0.7522	0.6372	0.2188	0.1107	0.9001	0.6298	0.6102	0.6655	-	-
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.5231	0.7143	0.9014	0.4286	0.7715	0.7226	0.3299	0.4708	0.8524	0.6054	0.6183	0.6787	-	-
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.4571	0.4062	0.898	0.3871	0.6625	0.7414	0.3211	0.5313	0.9137	0.6344	0.6154	0.6933	0.6500	-
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.4569	0.4308	0.8914	0.1724	0.7337	0.7499	0.3391	0.4119	0.945	0.6274	0.5814	0.6342	0.5294	-
completed	SeaFormer	SeaFormer-Base	ce+dice	0.4036	-	0.8814	-	0.7511	0.741	0.1886	0.2976	0.9058	0.6326	0.622	0.6428	-	-
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.3734	0.5000	0.889	-	0.7647	0.6167	0.1922	0.1699	0.884	0.629	0.6173	0.653	-	-
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.2443	0.4167	0.7843	-	0.6852	0.5775	0.1183	0.2243	0.9121	0.527	0.5869	0.6152	-	-

Remote-sensing-specific methods

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	FarSeg	ResNet-50	ce (native)	0.5052	-	0.7981	-	0.6712	0.6805	0.0227	0.4498	0.8901	0.6044	0.6624	0.6325	-	-
completed	FarSeg	ResNet-50	ce+dice	0.5878	-	0.7967	-	0.7369	0.5434	0.1559	0.3879	0.8669	0.5268	0.6511	0.4991	-	-
completed	FarSeg	ResNet-50	weighted_ce+dice	0.3804	0.2636	0.815	0.1818	0.6136	0.5931	0.3319	0.3801	0.8384	0.5011	0.6092	0.6596	0.5957	-
completed	FarSeg	ResNet-50	focal+dice	0.5183	-	0.8852	-	0.7195	0.7351	0.4562	0.5383	0.8962	0.5853	0.6343	0.6351	-	-
completed	BANet	ResT-Lite	ce+dice	0.3636	-	0.8668	-	0.7578	0.6915	0.3644	0.3926	0.8754	0.6199	0.6245	0.6488	-	-
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.4094	-	0.8464	-	0.746	0.7142	0.2931	0.4317	0.9321	0.6281	0.6034	0.6994	-	-
completed	MANet	ResNet-50	ce+dice	0.4241	-	0.8278	-	0.6036	0.6749	0.3082	0.4159	0.6822	0.5816	0.6015	0.6084	-	-
completed	MANet	ResNet-50	weighted_ce+dice	0.5852	0.1756	0.8348	0.0684	0.5345	0.6408	0.3270	0.3968	0.6498	0.6024	0.6138	0.6341	0.1228	-
completed	MANet	ResNet-50	focal+dice	0.4543	0.0789	0.8411	0.0670	0.5557	0.6537	0.2918	0.2626	0.3297	0.5715	0.5943	0.4826	0.2314	-
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.5180	-	0.9015	-	0.6625	0.7730	0.1986	0.4437	0.8679	0.6589	0.6733	0.6435	-	-
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.2201	0.3711	0.8839	-	0.6551	0.7447	0.2963	0.2839	0.4109	0.6099	0.6329	0.6542	-	-
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.4903	-	0.8776	-	0.6777	0.7665	0.2268	0.4301	0.415	0.624	0.6769	0.615	-	-
completed	DC-Swin	Swin-Small	ce+dice	0.3791	-	0.8698	-	0.7311	0.6213	0.3001	0.4737	0.9555	0.6131	0.6476	0.6384	-	-
completed	A2FPN	ResNet-18	ce+dice	0.4603	-	0.8091	-	0.6557	0.7575	0.2202	0.2621	0.9289	0.5628	0.5761	0.6699	-	-
completed	A2FPN	ResNet-18	weighted_ce+dice	0.4815	0.3226	0.8031	0.2632	0.6081	0.7581	0.4921	0.4350	0.9099	0.6085	0.5577	0.6761	0.3279	-
completed	A2FPN	ResNet-18	focal+dice	0.5574	-	0.8511	-	0.6407	0.6485	0.1781	0.2928	0.5652	0.5707	0.6095	0.6211	-	-
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.4412	0.1739	0.8913	0.4615	0.5881	0.7217	0.3223	0.2479	0.6975	0.5787	0.5816	0.6101	-	-
completed	FarSeg++	MiT-B2	ce (native)	0.4466	0.3810	0.8985	0.2105	0.7471	0.7235	0.1836	0.5502	0.9395	0.6217	0.6208	0.6713	0.1000	-
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.5269	-	0.8285	-	0.7178	0.7770	0.0609	0.4921	0.8392	0.6331	0.5882	0.6089	-	-
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.4731	0.1000	0.8265	-	0.6453	0.6851	-	0.4083	0.833	0.5837	0.5979	0.5805	-	-
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.4853	-	0.8836	-	0.6643	0.7055	0.2178	0.4489	0.7835	0.6215	0.6419	0.6718	-	-
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.4597	0.2692	0.8844	0.2449	0.5654	0.6049	0.3303	0.4172	0.5805	0.4909	0.6294	0.6097	0.6829	-
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.5315	-	0.8974	-	0.6423	0.7735	0.0946	0.4617	0.6539	0.6329	0.6595	0.6665	-	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	-	-	0.8935	-	0.4513	-	0.0956	-	0.2408	0.5715	0.5113	0.4592	-	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.1204	-	0.8944	-	0.6498	0.4819	0.3202	0.0873	0.307	0.5124	0.624	0.3356	-	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	-	-	0.8919	-	0.624	0.5105	0.2524	0.1625	0.2518	0.5835	0.6006	0.5089	-	-
completed	PyramidMamba	Swin-Base	ce+dice	0.5441	0.4400	0.8764	0.3750	0.7083	0.7235	0.2827	0.5512	0.8681	0.6515	0.6431	0.6644	-	-
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.5364	0.4337	0.8892	0.2927	0.6898	0.5839	0.2981	0.4072	0.8545	0.6451	0.6469	0.6855	0.5789	-
completed	PyramidMamba	Swin-Base	focal+dice	0.3980	0.0758	0.8678	0.1026	0.7325	0.7499	0.2271	0.4285	0.8919	0.6459	0.6219	0.5817	-	-
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.3046	-	0.8596	-	0.6569	0.4605	0.0947	0.3136	0.8584	0.5146	0.5954	0.6097	-	-
completed	MF-Mamba	HRNet-W18	ce+dice	0.5243	-	0.8055	-	0.7023	0.6695	0.0838	0.3843	0.9028	0.5088	0.6253	0.586	-	-
completed	MCPNet	ResNet-50	ce+dice	0.3089	-	0.8627	-	0.7178	0.7390	0.0670	0.4466	0.9411	0.6164	0.5726	0.6022	-	-
completed	MCPNet	ResNet-50	weighted_ce+dice	0.3253	0.1504	0.8983	-	0.7374	0.6799	0.2476	0.3955	0.9197	0.6682	0.6575	0.656	-	-
completed	MCPNet	ResNet-50	focal+dice	0.2126	-	0.8953	-	0.7477	0.6977	0.2199	0.3462	0.9237	0.5913	0.6496	0.6456	-	-

Methods related to vision foundation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.2234	0.1000	0.8909	-	0.556	0.2129	0.1615	0.1745	0.4572	0.407	0.4799	0.5076	-	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.1952	0.0865	0.8986	0.0323	0.5326	0.1707	0.3539	0.2016	0.3621	0.3998	0.4769	0.5282	-	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.381	0.1395	0.8961	-	0.5402	0.2425	0.2444	0.1665	0.4758	0.4204	0.4923	0.499	-	-
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.4211	-	0.8427	-	0.753	0.7662	0.373	0.2809	0.9461	0.6284	0.57	0.626	-	-
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.4678	-	0.828	-	0.7284	0.7636	0.1528	0.4729	0.9222	0.5969	0.5636	0.6387	-	-
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.2917	-	0.8848	-	0.6669	0.7090	0.1083	0.5108	0.9051	0.5669	0.6086	0.6186	-	-
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.4651	-	0.8387	-	0.6087	0.6658	0.1117	0.5358	0.7202	0.5609	0.6323	0.5875	-	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.2698	-	0.8882	-	0.6194	0.3161	0.1497	0.2358	0.4857	0.431	0.5582	0.5534	-	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.1001	0.0648	0.8955	0.0277	0.6388	0.2294	0.3437	0.1913	0.4787	0.4842	0.584	0.5662	-	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.2012	0.1500	0.8908	-	0.6679	0.3098	0.2818	0.2749	0.6992	0.4439	0.5622	0.5514	-	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.4057	-	0.837	-	0.7636	0.5984	0.1298	0.4166	0.8755	0.5882	0.6469	0.6648	-	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.4068	0.3368	0.8894	0.0151	0.6582	0.5738	0.1281	0.3383	0.7999	0.5993	0.6164	0.6283	-	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.4234	-	0.8882	-	0.7237	0.5802	0.0823	0.4945	0.9333	0.5129	0.565	0.6602	-	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.3285	-	0.8978	-	0.7623	0.5005	0.2108	0.3188	0.7911	0.594	0.5926	0.6136	-	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.5227	0.3333	0.8815	0.1224	0.7308	0.7305	0.2791	0.4422	0.7843	0.5967	0.5464	0.5997	0.3889	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.4452	-	0.8992	-	0.76	0.6685	0.1642	0.4000	0.8660	0.6208	0.5742	0.6515	-	-
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.4603	-	0.8091	-	0.6557	0.7575	0.2202	0.2621	0.9289	0.5628	0.5761	0.6699	-	-
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.5574	-	0.8511	-	0.6404	0.6485	0.1781	0.2928	0.5652	0.5707	0.6095	0.6211	-	-
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.4815	0.3226	0.8031	0.2632	0.608	0.7581	0.4921	0.4350	0.9099	0.6085	0.5577	0.6761	0.3279	-
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.4094	-	0.8464	-	0.746	0.7142	0.2931	0.4317	0.9321	0.6281	0.6034	0.6994	-	-
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.3636	-	0.8668	-	0.7578	0.6915	0.3644	0.3926	0.8754	0.6199	0.6245	0.6488	-	-
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.4241	-	0.8278	-	0.6036	0.6749	0.3082	0.4159	0.6822	0.5816	0.6015	0.6084	-	-
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.8966	-	0.7658	0.4000	0.5744	-	0.4	-	-	0.9326	0.4732	0.652	-	-
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.5852	0.1756	0.8348	0.0684	0.5345	0.6408	0.327	0.3968	0.6498	0.6024	0.6138	0.6341	0.1228	-
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.518	-	0.9015	-	0.6625	0.7730	0.1986	0.4437	0.8679	0.6589	0.6733	0.6434	-	-
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.4903	-	0.8776	-	0.6775	0.7665	0.2268	0.4301	0.4150	0.624	0.6769	0.6152	-	-
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.2201	0.3711	0.8839	-	0.6552	0.7447	0.2963	0.2839	0.4109	0.6099	0.6329	0.6542	-	-

Validation per-class ap

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/val_multilabel_per_class_ap.csv

General segmentation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.2787	0.588	0.9627	0.6703	0.8302	0.7679	0.1647	0.3514	0.9733	0.6421	0.5434	0.6982	0.195	-
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.3559	0.4668	0.947	0.6703	0.8288	0.7248	0.2134	0.5592	0.9735	0.6624	0.4707	0.6346	0.5471	-
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.2922	0.4504	0.9583	0.6703	0.8166	0.729	0.1713	0.4147	0.9706	0.629	0.5078	0.7028	0.0735	-
completed	DeepLabV3+	ResNet-50	ce+dice	0.4265	0.0009	0.9396	0.6072	0.7658	0.7228	0.321	0.4287	0.9715	0.6354	0.4735	0.5497	0.0009	-
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.379	0.67	0.9521	0.6704	0.7178	0.6955	0.206	0.376	0.9702	0.6017	0.4363	0.5428	0.0009	-
completed	DeepLabV3+	ResNet-50	focal+dice	0.3283	0.3795	0.9597	0.5048	0.7828	0.7921	0.2308	0.4583	0.9794	0.6217	0.4626	0.6098	0.0009	-
completed	UPerNet	Swin-Tiny	ce+dice	0.2678	0.1536	0.9348	0.0069	0.7872	0.4412	0.2018	0.2778	0.9538	0.5968	0.4538	0.5977	0.0009	-
completed	OCRNet	HRNet-W48	ce+dice	0.1909	0.0009	0.9339	0.0069	0.7665	0.4519	0.3283	0.3971	0.8683	0.5516	0.4263	0.6499	0.0009	-
completed	SegFormer	MiT-B2	ce+dice	0.3136	0.0746	0.9641	0.2149	0.7826	0.7565	0.3094	0.4961	0.9598	0.5749	0.5534	0.5882	0.1291	-
completed	SegFormer	MiT-B2	weighted_ce+dice	0.4016	0.6798	0.9504	0.0068	0.8421	0.4742	0.2915	0.5861	0.9072	0.6416	0.4155	0.6646	0.0009	-
completed	SegFormer	MiT-B2	focal+dice	0.3915	0.3967	0.9585	0.5765	0.8198	0.7032	0.2858	0.5292	0.9729	0.5917	0.4941	0.6112	0.0736	-
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.2732	0.2249	0.926	0.0865	0.6265	0.633	0.1966	0.2341	0.8217	0.3619	0.4427	0.6052	0.0009	-
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.2142	0.1675	0.9524	0.0069	0.8096	0.5143	0.1959	0.0767	0.9476	0.5766	0.5412	0.6225	0.0009	-
completed	Afformer	AFFormer-Base	ce+dice	0.2544	0.0607	0.9528	0.0289	0.8242	0.6431	0.1653	0.0536	0.9549	0.6105	0.5013	0.6377	0.0009	-
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.3919	0.5561	0.9574	0.3395	0.8161	0.7152	0.2868	0.4594	0.8845	0.5588	0.5081	0.6595	0.0009	-
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.3857	0.6003	0.9593	0.6704	0.7164	0.7658	0.2769	0.4527	0.9497	0.5952	0.4779	0.6407	0.6799	-
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.253	0.5426	0.9578	0.2858	0.8058	0.7219	0.2959	0.4899	0.9659	0.57	0.4587	0.5675	0.4794	-
completed	SeaFormer	SeaFormer-Base	ce+dice	0.1931	0.0009	0.9522	0.0069	0.8061	0.6716	0.1589	0.163	0.9447	0.6248	0.5275	0.6363	0.0009	-
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.2507	0.3341	0.9546	0.0069	0.8111	0.5991	0.199	0.0892	0.9585	0.5656	0.5051	0.6316	0.0009	-
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.1588	0.198	0.9286	0.0069	0.6252	0.4386	0.1754	0.1204	0.8543	0.3473	0.4242	0.5279	0.0009	-

Remote-sensing-specific methods

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	FarSeg	ResNet-50	ce (native)	0.3088	0.0009	0.926	0.0069	0.7862	0.6503	0.1293	0.4025	0.8162	0.5699	0.4912	0.6581	0.0009	-
completed	FarSeg	ResNet-50	ce+dice	0.4452	0.0009	0.9292	0.0069	0.8073	0.6981	0.1721	0.4568	0.93	0.4753	0.4757	0.6319	0.0009	-
completed	FarSeg	ResNet-50	weighted_ce+dice	0.3585	0.253	0.9332	0.6702	0.5541	0.6985	0.3184	0.5678	0.9663	0.4601	0.4332	0.5646	0.6874	-
completed	FarSeg	ResNet-50	focal+dice	0.3562	0.0009	0.95	0.0069	0.8106	0.6696	0.3452	0.4538	0.9028	0.5529	0.4755	0.6762	0.0009	-
completed	BANet	ResT-Lite	ce+dice	0.2492	0.0009	0.9455	0.0069	0.8004	0.6104	0.2967	0.3143	0.9412	0.5562	0.4657	0.6114	0.0009	-
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.2048	0.0009	0.9446	0.0069	0.7503	0.5947	0.2537	0.2819	0.9401	0.5363	0.4638	0.5839	0.0009	-
completed	MANet	ResNet-50	ce+dice	0.3683	0.0009	0.9285	0.0069	0.8228	0.7274	0.2127	0.4784	0.9546	0.515	0.4465	0.6634	0.0009	-
completed	MANet	ResNet-50	weighted_ce+dice	0.4677	0.7442	0.9258	0.3874	0.7713	0.7614	0.2653	0.5622	0.9629	0.5126	0.4577	0.6071	0.0413	-
completed	MANet	ResNet-50	focal+dice	0.3643	0.5207	0.938	0.2059	0.7565	0.7286	0.2493	0.5255	0.9321	0.5996	0.4764	0.6163	0.5472	-
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.3965	0.0009	0.96	0.0069	0.7767	0.685	0.1654	0.4747	0.9395	0.6423	0.5786	0.6872	0.0009	-
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.4834	0.4552	0.9581	0.0067	0.7463	0.7517	0.2065	0.5254	0.9577	0.5889	0.5017	0.6464	0.0009	-
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.349	0.0009	0.9593	0.0069	0.7906	0.7416	0.153	0.51	0.954	0.5834	0.5268	0.6911	0.0009	-
completed	DC-Swin	Swin-Small	ce+dice	0.225	0.0009	0.9427	0.0069	0.8169	0.5379	0.2243	0.4091	0.9533	0.5333	0.4826	0.6275	0.0009	-
completed	A2FPN	ResNet-18	ce+dice	0.3023	0.0009	0.9297	0.0069	0.6809	0.6815	0.2071	0.3753	0.9558	0.4783	0.4177	0.6336	0.0009	-
completed	A2FPN	ResNet-18	weighted_ce+dice	0.3312	0.2015	0.9259	0.1437	0.4891	0.7243	0.381	0.437	0.9461	0.5163	0.4051	0.5961	0.3884	-
completed	A2FPN	ResNet-18	focal+dice	0.397	0.0009	0.9479	0.0069	0.6322	0.6911	0.1684	0.477	0.9406	0.5326	0.4662	0.6303	0.0009	-
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.2253	0.0256	0.9424	0.2192	0.4791	0.7185	0.2518	0.1409	0.5914	0.4762	0.494	0.6375	0.0009	-
completed	FarSeg++	MiT-B2	ce (native)	0.2456	0.2879	0.9587	0.1066	0.7876	0.6394	0.2091	0.446	0.961	0.5731	0.502	0.6135	0.0286	-
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.346	0.0009	0.9375	0.0069	0.7183	0.7142	0.1438	0.4123	0.9684	0.6049	0.4392	0.5855	0.0009	-
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.3044	0.0564	0.931	0.0069	0.7167	0.6432	0.1194	0.2676	0.9394	0.4857	0.4257	0.5369	0.0009	-
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.3026	0.0009	0.9615	0.0069	0.7657	0.7335	0.2139	0.4844	0.9597	0.5468	0.5044	0.6085	0.0009	-
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.3697	0.2641	0.9554	0.4862	0.662	0.7543	0.2551	0.487	0.9666	0.4096	0.4838	0.5568	0.7396	-
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.3687	0.0009	0.9604	0.0069	0.8035	0.6747	0.162	0.5771	0.9479	0.6466	0.5418	0.6356	0.0009	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0.0505	0.0009	0.9499	0.0069	0.4991	0.0337	0.1097	0.0154	0.7364	0.5284	0.4166	0.6607	0.0009	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.1796	0.0009	0.9618	0.0069	0.8249	0.6256	0.2097	0.2387	0.8575	0.5236	0.4986	0.5871	0.0009	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0.0505	0.0009	0.9595	0.0069	0.7514	0.3802	0.1361	0.0632	0.8005	0.5519	0.5034	0.6915	0.0009	-
completed	PyramidMamba	Swin-Base	ce+dice	0.4059	0.282	0.9538	0.6704	0.8314	0.7644	0.2525	0.5725	0.9696	0.6686	0.4998	0.6081	0.0009	-
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.4065	0.5433	0.9636	0.6703	0.8391	0.7903	0.2518	0.5966	0.9727	0.6522	0.5132	0.6788	0.5997	-
completed	PyramidMamba	Swin-Base	focal+dice	0.4245	0.0901	0.9532	0.4516	0.8185	0.7598	0.2444	0.6132	0.9603	0.6368	0.4592	0.6648	0.0009	-
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.1301	0.0009	0.9376	0.0069	0.7189	0.3704	0.1218	0.1559	0.9103	0.4412	0.4292	0.5421	0.0009	-
completed	MF-Mamba	HRNet-W18	ce+dice	0.3348	0.0009	0.9367	0.0069	0.7266	0.7039	0.1643	0.4693	0.957	0.4238	0.4485	0.5596	0.0009	-
completed	MCPNet	ResNet-50	ce+dice	0.1299	0.0009	0.9479	0.0067	0.7641	0.6639	0.1456	0.3291	0.9649	0.542	0.453	0.5355	0.0009	-
completed	MCPNet	ResNet-50	weighted_ce+dice	0.1489	0.2011	0.9562	0.0062	0.7877	0.5851	0.2054	0.2762	0.9335	0.636	0.5548	0.608	0.0009	-
completed	MCPNet	ResNet-50	focal+dice	0.1273	0.0009	0.9465	0.0069	0.7914	0.5877	0.209	0.2344	0.8896	0.5519	0.5604	0.5592	0.0009	-

Methods related to vision foundation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.2191	0.0096	0.9383	0.0069	0.8205	0.5164	0.1127	0.0951	0.9262	0.5653	0.461	0.626	0.0009	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.1944	0.0208	0.9429	0.3143	0.7978	0.4582	0.2012	0.0800	0.9001	0.5689	0.4237	0.6063	0.0009	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.3224	0.0276	0.9435	0.0069	0.8053	0.5270	0.1235	0.0614	0.9164	0.5613	0.4695	0.6518	0.0009	-
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.2031	0.0009	0.9465	0.0069	0.7645	0.6529	0.2829	0.1778	0.9155	0.5509	0.4488	0.5705	0.0009	-
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.2971	0.0009	0.9396	0.0069	0.7492	0.6576	0.1918	0.3237	0.8937	0.5026	0.3845	0.552	0.0009	-
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.1807	0.0009	0.9488	0.0069	0.6934	0.6238	0.181	0.3679	0.9455	0.5203	0.4782	0.5123	0.0009	-
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.2762	0.0009	0.9334	0.0069	0.6358	0.7504	0.1578	0.4492	0.9548	0.535	0.4824	0.5119	0.0009	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.2528	0.0009	0.947	0.0069	0.7846	0.4317	0.1195	0.2352	0.9104	0.5679	0.4376	0.5516	0.0009	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.0876	0.0913	0.9573	0.1113	0.7674	0.5730	0.205	0.1898	0.9350	0.6007	0.4587	0.5404	0.0009	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.1419	0.0205	0.9528	0.0069	0.7885	0.5741	0.1858	0.2644	0.9339	0.5852	0.4557	0.5367	0.0009	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.2712	0.0009	0.9355	0.0069	0.8153	0.4620	0.1548	0.5399	0.9088	0.5377	0.4777	0.5982	0.0009	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.2498	0.4262	0.953	0.0147	0.7485	0.5467	0.1779	0.4906	0.9318	0.5421	0.4958	0.5461	0.0009	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.2932	0.0009	0.9518	0.0069	0.7747	0.5594	0.1462	0.4964	0.9502	0.4687	0.4718	0.5829	0.0009	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.3176	0.0009	0.9518	0.0069	0.8274	0.4850	0.1898	0.4126	0.9261	0.576	0.4695	0.6444	0.0009	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.4232	0.5554	0.9507	0.2181	0.8112	0.6726	0.1887	0.5730	0.9518	0.5181	0.4015	0.6179	0.5650	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.3914	0.0009	0.9553	0.0069	0.8028	0.6032	0.1957	0.2646	0.9265	0.5695	0.5141	0.6546	0.0009	-
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.3024	0.0009	0.9297	0.0069	0.682	0.6814	0.2076	0.3745	0.9558	0.4783	0.4177	0.6337	0.0009	-
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.399	0.0009	0.9479	0.0069	0.6335	0.6909	0.1685	0.4685	0.9400	0.5327	0.4665	0.6298	0.0009	-
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.3311	0.2015	0.9259	0.1437	0.4898	0.7242	0.3809	0.4366	0.9460	0.5164	0.4052	0.5957	0.3914	-
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.2059	0.0009	0.9447	0.0069	0.751	0.5947	0.2537	0.2804	0.9399	0.5364	0.4639	0.5836	0.0009	-
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.2503	0.0009	0.9455	0.0069	0.8007	0.6104	0.2966	0.3153	0.9411	0.5561	0.4659	0.6112	0.0009	-
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.3704	0.0009	0.9285	0.0069	0.8234	0.7275	0.2129	0.4812	0.9546	0.515	0.4467	0.6631	0.0009	-
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.8448	-	0.9281	0.6017	0.8329	-	0.3077	-	-	0.9305	0.4889	0.8205	-	-
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.469	0.7479	0.9258	0.3874	0.7718	0.7614	0.2654	0.5617	0.9629	0.5125	0.4578	0.6069	0.0413	-
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.395	0.0009	0.9601	0.0069	0.7797	0.6850	0.1662	0.4741	0.9393	0.6425	0.579	0.6866	0.0009	-
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.3487	0.0009	0.9593	0.0069	0.7939	0.7416	0.1531	0.5089	0.9538	0.5833	0.5272	0.6908	0.0009	-
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.4842	0.4589	0.9581	0.0067	0.7518	0.7517	0.2067	0.5239	0.9575	0.5892	0.5018	0.6461	0.0009	-

Test

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/test_multilabel_summary.csv

General segmentation models

status	model	backbone	loss	cp	cr	cf1	op	or	of1	map	macro_f1	micro_f1	sample_f1
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.5586	0.6951	0.6194	0.685	0.8133	0.7437	0.6138	0.6135	0.7437	0.7574
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.55	0.6924	0.6131	0.6732	0.8044	0.733	0.5937	0.6015	0.733	0.7395
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.5536	0.6728	0.6074	0.6665	0.8223	0.7362	0.5926	0.5954	0.7362	0.7458
completed	DeepLabV3+	ResNet-50	ce+dice	0.4948	0.6274	0.5533	0.6781	0.8083	0.7375	0.5582	0.6833	0.7375	0.7561
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.5404	0.6461	0.5886	0.6803	0.7885	0.7304	0.5584	0.6417	0.7304	0.7447
completed	DeepLabV3+	ResNet-50	focal+dice	0.4881	0.714	0.5798	0.6325	0.8604	0.7291	0.5796	0.6257	0.7291	0.7424
completed	UPerNet	Swin-Tiny	ce+dice	0.5441	0.5653	0.5545	0.704	0.7645	0.733	0.5011	0.6095	0.733	0.7422
completed	OCRNet	HRNet-W48	ce+dice	0.5009	0.5395	0.5195	0.6579	0.7061	0.6811	0.4633	0.62	0.6811	0.6589
completed	SegFormer	MiT-B2	ce+dice	0.5309	0.6277	0.5753	0.6929	0.7207	0.7065	0.5235	0.5597	0.7065	0.7045
completed	SegFormer	MiT-B2	weighted_ce+dice	0.5665	0.6796	0.6179	0.734	0.775	0.7539	0.6167	0.6747	0.7539	0.779
completed	SegFormer	MiT-B2	focal+dice	0.5288	0.6958	0.6009	0.6582	0.7594	0.7052	0.5691	0.5735	0.7052	0.6986
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.6584	0.5126	0.5764	0.763	0.6625	0.7092	0.5094	0.6317	0.7092	0.7125
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.5306	0.5334	0.532	0.6659	0.7484	0.7048	0.4682	0.5636	0.7048	0.6888
completed	Afformer	AFFormer-Base	ce+dice	0.522	0.6169	0.5655	0.6572	0.7639	0.7065	0.5154	0.5497	0.7065	0.6983
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.5583	0.6376	0.5953	0.7063	0.7937	0.7474	0.5625	0.624	0.7474	0.7711
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.5896	0.6812	0.6321	0.7031	0.8023	0.7494	0.5904	0.6113	0.7494	0.7683
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.5822	0.6753	0.6253	0.684	0.7881	0.7324	0.5776	0.6056	0.7324	0.7375
completed	SeaFormer	SeaFormer-Base	ce+dice	0.5405	0.577	0.5582	0.6775	0.7554	0.7143	0.4942	0.5941	0.7143	0.7064
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.586	0.5601	0.5728	0.6472	0.7446	0.6925	0.482	0.5247	0.6925	0.6736
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.6373	0.4949	0.5571	0.7678	0.6664	0.7135	0.4868	0.6062	0.7135	0.711

Remote-sensing-specific methods

status	model	backbone	loss	cp	cr	cf1	op	or	of1	map	macro_f1	micro_f1	sample_f1
completed	FarSeg	ResNet-50	ce (native)	0.488	0.6234	0.5475	0.7013	0.8159	0.7543	0.5263	0.6695	0.7543	0.7658
completed	FarSeg	ResNet-50	ce+dice	0.4617	0.643	0.5375	0.6716	0.8039	0.7318	0.5562	0.6499	0.7318	0.742
completed	FarSeg	ResNet-50	weighted_ce+dice	0.4766	0.7369	0.5788	0.6265	0.7857	0.6971	0.5762	0.5455	0.6971	0.705
completed	FarSeg	ResNet-50	focal+dice	0.4884	0.6471	0.5567	0.6889	0.8537	0.7625	0.5582	0.6822	0.7625	0.7887
completed	BANet	ResT-Lite	ce+dice	0.4616	0.5574	0.505	0.6664	0.7606	0.7104	0.4664	0.6238	0.7104	0.7155
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.5194	0.5399	0.5294	0.7339	0.7366	0.7353	0.4916	0.6593	0.7353	0.746
completed	MANet	ResNet-50	ce+dice	0.4338	0.6937	0.5338	0.6062	0.8818	0.7185	0.5693	0.6545	0.7185	0.7376
completed	MANet	ResNet-50	weighted_ce+dice	0.4458	0.8443	0.5835	0.586	0.8794	0.7034	0.6559	0.5535	0.7034	0.7236
completed	MANet	ResNet-50	focal+dice	0.4344	0.861	0.5774	0.5777	0.8897	0.7005	0.6643	0.5444	0.7005	0.7203
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.5221	0.639	0.5747	0.6912	0.8521	0.7633	0.5648	0.7105	0.7633	0.7839
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.4444	0.7464	0.5571	0.6076	0.8561	0.7107	0.5826	0.5893	0.7107	0.7314
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.4633	0.6671	0.5469	0.6344	0.8535	0.7278	0.5613	0.6704	0.7278	0.7533
completed	DC-Swin	Swin-Small	ce+dice	0.4535	0.5773	0.508	0.6489	0.8161	0.723	0.4726	0.6255	0.723	0.7345
completed	A2FPN	ResNet-18	ce+dice	0.5133	0.5665	0.5386	0.7381	0.7526	0.7453	0.5239	0.6678	0.7453	0.7725
completed	A2FPN	ResNet-18	weighted_ce+dice	0.566	0.6809	0.6182	0.7081	0.7544	0.7305	0.5982	0.6064	0.7305	0.7554
completed	A2FPN	ResNet-18	focal+dice	0.4574	0.623	0.5275	0.6502	0.7846	0.7111	0.5219	0.6394	0.7111	0.7161
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.5961	0.5941	0.5951	0.6678	0.8413	0.7446	0.5335	0.6306	0.7446	0.7664
completed	FarSeg++	MiT-B2	ce (native)	0.54	0.5397	0.5399	0.7078	0.7409	0.724	0.4864	0.5932	0.724	0.726
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.62	0.5916	0.6055	0.68	0.7801	0.7266	0.5281	0.6192	0.7266	0.7213
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.6101	0.5911	0.6004	0.7001	0.7815	0.7385	0.5145	0.6207	0.7385	0.7455
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.4914	0.6305	0.5523	0.667	0.8063	0.7301	0.5402	0.6806	0.7301	0.7332
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.5786	0.7293	0.6453	0.6509	0.8175	0.7247	0.622	0.6273	0.7247	0.7318
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.4956	0.6459	0.5609	0.6726	0.8394	0.7468	0.5624	0.6895	0.7468	0.7636
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0.3354	0.4376	0.3797	0.6214	0.8425	0.7153	0.4047	0.753	0.7153	0.749
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.3574	0.692	0.4714	0.5223	0.8905	0.6584	0.4716	0.5523	0.6584	0.6762
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0.4559	0.5104	0.4816	0.6167	0.872	0.7225	0.4554	0.6587	0.7225	0.7541
completed	PyramidMamba	Swin-Base	ce+dice	0.5523	0.6965	0.6161	0.6568	0.8324	0.7343	0.605	0.668	0.7343	0.7337
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.5602	0.7758	0.6506	0.6263	0.8459	0.7197	0.6749	0.6181	0.7197	0.7357
completed	PyramidMamba	Swin-Base	focal+dice	0.4768	0.7094	0.5703	0.6385	0.8289	0.7213	0.5994	0.6183	0.7213	0.7184
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.5133	0.5051	0.5091	0.6588	0.746	0.6997	0.419	0.5142	0.6997	0.6945
completed	MF-Mamba	HRNet-W18	ce+dice	0.4403	0.6114	0.5119	0.6508	0.789	0.7133	0.4807	0.6195	0.7133	0.7104
completed	MCPNet	ResNet-50	ce+dice	0.4849	0.5565	0.5182	0.6956	0.7545	0.7238	0.4649	0.6357	0.7238	0.7242
completed	MCPNet	ResNet-50	weighted_ce+dice	0.5195	0.6184	0.5646	0.6749	0.7615	0.7156	0.5141	0.616	0.7156	0.713
completed	MCPNet	ResNet-50	focal+dice	0.4945	0.5715	0.5302	0.6993	0.7839	0.7392	0.4849	0.5524	0.7392	0.7538

Methods related to vision foundation models

status	model	backbone	loss	cp	cr	cf1	op	or	of1	map	macro_f1	micro_f1	sample_f1
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.3647	0.6824	0.4754	0.4954	0.9257	0.6455	0.4749	0.5098	0.6455	0.6488
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.328	0.737	0.454	0.4455	0.9243	0.6012	0.4625	0.4226	0.6012	0.5995
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.3846	0.656	0.4849	0.5066	0.9178	0.6529	0.4821	0.5258	0.6529	0.6623
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.53	0.5205	0.5252	0.7355	0.7214	0.7284	0.4666	0.6442	0.7284	0.7352
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.5355	0.5312	0.5334	0.7243	0.7374	0.7308	0.4615	0.6571	0.7308	0.7269
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.4921	0.5071	0.4995	0.7213	0.7331	0.7271	0.4528	0.6172	0.7271	0.7409
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.4799	0.5933	0.5306	0.6829	0.7843	0.7301	0.4944	0.6442	0.7301	0.7384
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.3745	0.6554	0.4767	0.5165	0.906	0.6579	0.4621	0.5117	0.6579	0.6603
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.3157	0.7515	0.4446	0.4238	0.9081	0.5779	0.4107	0.4031	0.5779	0.5986
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.3508	0.6742	0.4615	0.4933	0.9156	0.6412	0.4541	0.4904	0.6412	0.6411
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.4654	0.5394	0.4996	0.6785	0.7464	0.7108	0.4595	0.6116	0.7108	0.7148
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.427	0.6527	0.5163	0.6359	0.7738	0.6981	0.485	0.5313	0.6981	0.7258
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.4484	0.577	0.5046	0.6897	0.7728	0.7289	0.464	0.6143	0.7289	0.744
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.4322	0.628	0.5121	0.6393	0.8147	0.7164	0.5046	0.6171	0.7164	0.74
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.4842	0.7614	0.592	0.6559	0.814	0.7264	0.5869	0.5666	0.7264	0.7512
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.5499	0.6354	0.5896	0.7046	0.823	0.7592	0.5454	0.6147	0.7592	0.7843
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.5133	0.5665	0.5386	0.7381	0.7526	0.7453	0.524	0.6678	0.7453	0.7725
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.4574	0.623	0.5275	0.6502	0.7846	0.7111	0.5218	0.6394	0.7111	0.7161
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.566	0.6809	0.6181	0.7081	0.7544	0.7305	0.5983	0.6064	0.7305	0.7553
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.5193	0.5399	0.5294	0.7338	0.7367	0.7353	0.4916	0.6593	0.7353	0.746
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.4616	0.5574	0.505	0.6664	0.7606	0.7104	0.4664	0.6238	0.7104	0.7155
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.4261	0.6917	0.5273	0.6085	0.8824	0.7203	0.5617	0.643	0.7203	0.734
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.4344	0.861	0.5774	0.5777	0.8897	0.7005	0.6645	0.5444	0.7005	0.7203
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.4458	0.8443	0.5835	0.586	0.8794	0.7033	0.6557	0.5535	0.7033	0.7236
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.5221	0.639	0.5747	0.6912	0.8521	0.7633	0.5646	0.7105	0.7633	0.7838
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.4633	0.6671	0.5468	0.6343	0.8535	0.7278	0.5612	0.6704	0.7278	0.7533
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.4444	0.7464	0.5571	0.6075	0.8561	0.7107	0.5828	0.5893	0.7107	0.7314

Test per-class precision

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/test_multilabel_per_class_precision.csv

General segmentation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.5162	0.474	0.841	0	0.8437	0.3058	0	0.5325	0	0.676	0.5129	0.6754	0.208	0
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.4673	0.5278	0.8399	0	0.863	0.2616	0	0.4521	0	0.641	0.4911	0.6873	0.2692	0
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.5323	0.3722	0.8191	0	0.8314	0.3472	0	0.5934	0	0.6496	0.4649	0.6981	0.2273	0
completed	DeepLabV3+	ResNet-50	ce+dice	0.5466	0	0.8235	0	0.861	0.3414	0	0.5099	0	0.716	0.4965	0.6528	0	0
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.5896	0.5431	0.8274	0	0.8317	0.1594	0	0.5829	0	0.6574	0.532	0.6807	0	0
completed	DeepLabV3+	ResNet-50	focal+dice	0.3256	0.4663	0.8131	0	0.8411	0.1739	0	0.5777	0	0.6171	0.4519	0.6147	0	0
completed	UPerNet	Swin-Tiny	ce+dice	0.4679	0.4416	0.8407	0	0.8658	0.4266	0	0.5405	0	0.6795	0.4947	0.6836	0	0
completed	OCRNet	HRNet-W48	ce+dice	0.6468	0	0.9082	0	0.8854	0.1501	0	0.6308	0	0.743	0.4625	0.5821	0	0
completed	SegFormer	MiT-B2	ce+dice	0.2513	0.2863	0.8927	0	0.8673	0.3553	0	0.5771	0	0.7042	0.5057	0.6728	0.1966	0
completed	SegFormer	MiT-B2	weighted_ce+dice	0.476	0.5427	0.8536	0	0.9064	0.4792	0	0.449	0	0.7138	0.4956	0.7487	0	0
completed	SegFormer	MiT-B2	focal+dice	0.1762	0.4561	0.8535	0	0.8728	0.3675	0	0.5313	0	0.7465	0.513	0.5976	0.1733	0
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.7366	0.7031	0.8756	0	0.9591	0.4711	0	0.8289	0	0.7561	0.5391	0.7144	0	0
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.4267	0.625	0.861	0	0.8329	0.2201	0	0.5853	0	0.6702	0.4765	0.6077	0	0
completed	Afformer	AFFormer-Base	ce+dice	0.3137	0.4048	0.8371	0	0.8305	0.2752	0	0.562	0	0.6669	0.5012	0.5983	0.2308	0
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.4416	0.75	0.8137	0	0.8976	0.2791	0	0.413	0	0.7105	0.5291	0.7489	0	0
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.3606	0.5773	0.8425	0	0.8479	0.2673	0	0.6083	0	0.6846	0.5197	0.7276	0.46	0
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.5079	0.4301	0.8379	0	0.8694	0.3458	0	0.4827	0	0.721	0.4592	0.684	0.4839	0
completed	SeaFormer	SeaFormer-Base	ce+dice	0.3571	0.4848	0.8401	0	0.8657	0.5458	0	0.5263	0	0.6943	0.5106	0.5806	0	0
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.3855	0.8529	0.856	0	0.8426	0.2702	0	0.6595	0	0.6179	0.4607	0.5809	0.3333	0
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.7288	0.4286	0.8872	0	0.9377	0.5601	0	0.8582	0	0.7178	0.5909	0.6637	0	0

Remote-sensing-specific methods

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	FarSeg	ResNet-50	ce (native)	0.5496	0	0.8472	0	0.8818	0.2501	0	0.4509	0	0.7119	0.4802	0.7084	0	0
completed	FarSeg	ResNet-50	ce+dice	0.4232	0	0.8604	0	0.8633	0.2351	0	0.3809	0	0.6679	0.4952	0.6906	0	0
completed	FarSeg	ResNet-50	weighted_ce+dice	0.1573	0.2269	0.8588	0	0.8411	0.2414	0	0.3712	0	0.5607	0.4718	0.675	0.3617	0
completed	FarSeg	ResNet-50	focal+dice	0.3912	0	0.8384	0	0.88	0.2341	0	0.7203	0	0.6183	0.5156	0.686	0	0
completed	BANet	ResT-Lite	ce+dice	0.3295	0	0.8389	0	0.8222	0.35	0	0.4527	0	0.6768	0.5114	0.6348	0	0
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.5744	0	0.876	0	0.9035	0.2948	0	0.6229	0	0.7017	0.5	0.7207	0	0
completed	MANet	ResNet-50	ce+dice	0.4338	0	0.7336	0	0.8	0.2406	0	0.4097	0	0.6418	0.4643	0.6146	0	0
completed	MANet	ResNet-50	weighted_ce+dice	0.3733	0.1513	0.7528	0	0.7911	0.1688	0	0.3331	0	0.6184	0.4654	0.6391	0.1649	0
completed	MANet	ResNet-50	focal+dice	0.244	0.1799	0.7769	0	0.8182	0.2	0	0.3526	0	0.5895	0.4553	0.5858	0.1417	0
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.5664	0	0.7863	0	0.8409	0.6527	0	0.4908	0	0.6953	0.5353	0.6537	0	0
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.2264	0.3356	0.84	0	0.8754	0.1943	0	0.2942	0	0.5882	0.4869	0.6031	0	0
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.3989	0	0.8331	0	0.8577	0.3754	0	0.3989	0	0.6343	0.5154	0.6196	0	0
completed	DC-Swin	Swin-Small	ce+dice	0.3918	0	0.8606	0	0.8389	0.3095	0	0.3737	0	0.6307	0.4923	0.6374	0	0
completed	A2FPN	ResNet-18	ce+dice	0.5446	0	0.8812	0	0.8829	0.3282	0	0.5233	0	0.6938	0.5248	0.7544	0	0
completed	A2FPN	ResNet-18	weighted_ce+dice	0.5229	0.4379	0.8741	0	0.89	0.343	0	0.4421	0	0.68	0.5329	0.7083	0.2289	0
completed	A2FPN	ResNet-18	focal+dice	0.5019	0	0.8541	0	0.8909	0.2329	0	0.3482	0	0.6271	0.4995	0.6198	0	0
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.6632	0.766	0.8097	0	0.816	0.5204	0	0.6066	0	0.6361	0.4874	0.6558	0	0
completed	FarSeg++	MiT-B2	ce (native)	0.426	0.4898	0.8502	0	0.8668	0.2955	0	0.567	0	0.7258	0.5411	0.6376	0	0
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.5491	1	0.8149	0	0.8983	0.4537	0	0.7018	0	0.6809	0.4864	0.6144	0	0
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.5137	0.931	0.7975	0	0.8481	0.2753	0	0.8514	0	0.7074	0.5082	0.6684	0	0
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.4713	0	0.8413	0	0.87	0.4266	0	0.5167	0	0.6812	0.48	0.6264	0	0
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.4103	0.5714	0.83	0	0.8531	0.2819	0	0.4813	0	0.6286	0.502	0.627	0.6	0
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.4567	0	0.7672	0	0.8491	0.631	0	0.4222	0	0.6447	0.5059	0.6795	0	0
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0	0	0.8031	0	0.8362	0	0	0	0	0.6668	0.4591	0.5885	0	0
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.0907	0	0.8324	0	0.7677	0.1718	0	0.3062	0	0.489	0.4499	0.4669	0	0
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0	0	0.7743	0	0.8241	0.4831	0	0.7765	0	0.654	0.4561	0.5911	0	0
completed	PyramidMamba	Swin-Base	ce+dice	0.4575	0.701	0.8275	0	0.8591	0.301	0	0.6159	0	0.6708	0.4586	0.6312	0	0
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.3395	0.6014	0.7747	0	0.8137	0.121	0	0.5529	0	0.6678	0.4902	0.6474	0.5932	0
completed	PyramidMamba	Swin-Base	focal+dice	0.3882	0.2601	0.7905	0	0.8854	0.2819	0	0.4705	0	0.6469	0.4848	0.5595	0	0
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.205	0.7273	0.8369	0	0.8065	0.1942	0	0.6101	0	0.6355	0.4995	0.6177	0	0
completed	MF-Mamba	HRNet-W18	ce+dice	0.3841	0	0.8396	0	0.8669	0.1704	0	0.4209	0	0.6324	0.4927	0.5958	0	0
completed	MCPNet	ResNet-50	ce+dice	0.3239	0	0.8424	0	0.8375	0.2887	0	0.6872	0	0.7326	0.5213	0.6152	0	0
completed	MCPNet	ResNet-50	weighted_ce+dice	0.3832	0.2249	0.8615	0	0.853	0.4669	0	0.5584	0	0.6994	0.517	0.6303	0	0
completed	MCPNet	ResNet-50	focal+dice	0.1524	0	0.833	0	0.8594	0.3507	0	0.5137	0	0.7168	0.518	0.6677	0.3333	0

Methods related to vision foundation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.1642	0.2628	0.7596	0	0.6915	0.0932	0	0.3299	0	0.4504	0.3656	0.5303	0	0
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.1097	0.0752	0.7819	0	0.7005	0.0484	0	0.2468	0	0.3989	0.378	0.5331	0.0078	0
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.2821	0.233	0.7806	0	0.7276	0.0956	0	0.3904	0	0.45	0.3719	0.5143	0	0
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.426	0	0.8795	0	0.8971	0.301	0	0.8786	0	0.7204	0.509	0.6879	0	0
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.5424	0	0.8576	0	0.897	0.3295	0	0.8533	0	0.6987	0.5005	0.6764	0	0
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.5833	0	0.8558	0	0.8432	0.232	0	0.4288	0	0.7416	0.5724	0.6635	0	0
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.3704	0	0.842	0	0.8471	0.1822	0	0.6568	0	0.7283	0.5458	0.6267	0	0
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.1826	0.2879	0.7338	0	0.7313	0.1422	0	0.254	0	0.4758	0.4148	0.5228	0	0
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.0425	0.0387	0.7292	0	0.7062	0.047	0	0.2082	0	0.4253	0.415	0.526	0.0185	0
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.0918	0.2706	0.7267	0	0.7112	0.0637	0	0.3047	0	0.4282	0.4006	0.5104	0	0
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.3894	0	0.8472	0	0.8654	0.3226	0	0.3635	0	0.7393	0.4823	0.6439	0	0
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.1896	0.1053	0.8089	0	0.8472	0.1454	0	0.2988	0	0.6884	0.5393	0.6472	0	0
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.2987	0	0.8236	0	0.8752	0.218	0	0.4111	0	0.6653	0.5497	0.6423	0	0
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.3065	0	0.7871	0	0.8696	0.2297	0	0.2873	0	0.6841	0.505	0.6531	0	0
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.382	0.1764	0.8387	0	0.8184	0.2743	0	0.3753	0	0.6305	0.5225	0.6682	0.1559	0
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.2515	0.7647	0.785	0	0.8595	0.4046	0	0.4709	0	0.7151	0.5746	0.6732	0	0
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.5446	0	0.8812	0	0.8828	0.3282	0	0.5233	0	0.6938	0.5248	0.7544	0	0
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.5019	0	0.8541	0	0.8905	0.2329	0	0.3482	0	0.6271	0.4995	0.6197	0	0
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.5229	0.4379	0.8741	0	0.8896	0.343	0	0.4421	0	0.68	0.5329	0.7083	0.2289	0
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.5744	0	0.876	0	0.903	0.2948	0	0.6229	0	0.7017	0.5	0.7207	0	0
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.3295	0	0.8389	0	0.8222	0.35	0	0.4527	0	0.6768	0.5114	0.6348	0	0
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.3796	0	0.7491	0	0.7905	0.1939	0	0.4111	0	0.6528	0.4595	0.6242	0	0
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.244	0.1799	0.7769	0	0.8181	0.2	0	0.3526	0	0.5895	0.4553	0.5858	0.1417	0
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.3733	0.1513	0.7528	0	0.7911	0.1688	0	0.3331	0	0.6184	0.4654	0.6391	0.1649	0
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.5664	0	0.7863	0	0.8405	0.6527	0	0.4908	0	0.6953	0.5353	0.6537	0	0
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.3989	0	0.8331	0	0.8574	0.3754	0	0.3989	0	0.6343	0.5154	0.6196	0	0
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.2264	0.3356	0.84	0	0.8748	0.1943	0	0.2942	0	0.5882	0.4869	0.6031	0	0

Test per-class recall

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/test_multilabel_per_class_recall.csv

General segmentation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.6991	0.5105	0.8098	-	0.8897	0.6228	-	0.7845	-	0.7338	0.7266	0.8978	0.2766	-
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.7398	0.3986	0.7558	-	0.8838	0.519	-	0.8492	-	0.808	0.7302	0.8674	0.3723	-
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.6708	0.5804	0.7978	-	0.9017	0.5731	-	0.7018	-	0.7754	0.7909	0.8824	0.0532	-
completed	DeepLabV3+	ResNet-50	ce+dice	0.8088	0	0.8154	-	0.8843	0.5789	-	0.8748	-	0.6937	0.7321	0.8862	0	-
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.7837	0.4406	0.8065	-	0.8662	0.5468	-	0.7778	-	0.8059	0.5931	0.8404	0	-
completed	DeepLabV3+	ResNet-50	focal+dice	0.8401	0.5315	0.8585	-	0.8988	0.6813	-	0.7623	-	0.8246	0.8162	0.9265	0	-
completed	UPerNet	Swin-Tiny	ce+dice	0.6176	0.2378	0.778	-	0.8653	0.3655	-	0.6023	-	0.7327	0.5888	0.8656	0	-
completed	OCRNet	HRNet-W48	ce+dice	0.511	0	0.5617	-	0.7444	0.5424	-	0.6674	-	0.6887	0.7618	0.918	0	-
completed	SegFormer	MiT-B2	ce+dice	0.7429	0.4825	0.69	-	0.774	0.4327	-	0.756	-	0.6961	0.6579	0.8006	0.2447	-
completed	SegFormer	MiT-B2	weighted_ce+dice	0.8997	0.6224	0.8214	-	0.8768	0.6067	-	0.9	-	0.8064	0.4391	0.8232	0	-
completed	SegFormer	MiT-B2	focal+dice	0.9248	0.5455	0.7707	-	0.7259	0.6287	-	0.7845	-	0.676	0.7327	0.8926	0.2766	-
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.5172	0.3147	0.77	-	0.5852	0.4284	-	0.5779	-	0.6516	0.5655	0.7158	0	-
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.6113	0.1399	0.6851	-	0.7739	0.1725	-	0.5317	-	0.7759	0.7466	0.8969	0	-
completed	Afformer	AFFormer-Base	ce+dice	0.7179	0.3566	0.7465	-	0.733	0.4708	-	0.6279	-	0.7984	0.7323	0.8897	0.0957	-
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.6991	0.2098	0.8483	-	0.8604	0.7193	-	0.8593	-	0.7197	0.6043	0.8556	0	-
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.7868	0.3916	0.8079	-	0.8945	0.6711	-	0.7232	-	0.7515	0.6857	0.8547	0.2447	-
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.6019	0.5804	0.7695	-	0.8539	0.6213	-	0.8597	-	0.71	0.7209	0.8759	0.1596	-
completed	SeaFormer	SeaFormer-Base	ce+dice	0.721	0.1119	0.7778	-	0.7156	0.4708	-	0.648	-	0.7834	0.631	0.9102	0	-
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.7022	0.2028	0.7142	-	0.7124	0.3611	-	0.4481	-	0.8303	0.702	0.8963	0.0319	-
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.4044	0.1678	0.7862	-	0.5615	0.5585	-	0.4956	-	0.7148	0.5308	0.7292	0	-

Remote-sensing-specific methods

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	FarSeg	ResNet-50	ce (native)	0.6426	0	0.767	-	0.905	0.6228	-	0.866	-	0.7343	0.7956	0.901	0	-
completed	FarSeg	ResNet-50	ce+dice	0.8119	0	0.6721	-	0.9158	0.6243	-	0.9	-	0.794	0.814	0.8981	0	-
completed	FarSeg	ResNet-50	weighted_ce+dice	0.8464	0.6014	0.6972	-	0.8938	0.7193	-	0.8984	-	0.7742	0.7152	0.8611	0.3617	-
completed	FarSeg	ResNet-50	focal+dice	0.8683	0	0.8557	-	0.9123	0.5702	-	0.7375	-	0.8854	0.7489	0.8927	0	-
completed	BANet	ResT-Lite	ce+dice	0.6364	0	0.7913	-	0.7884	0.4401	-	0.6888	-	0.7237	0.6612	0.8445	0	-
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.605	0	0.7599	-	0.8281	0.3991	-	0.743	-	0.7294	0.5424	0.7921	0	-
completed	MANet	ResNet-50	ce+dice	0.9342	0	0.8747	-	0.9362	0.693	-	0.8979	-	0.8368	0.8341	0.9303	0	-
completed	MANet	ResNet-50	weighted_ce+dice	0.9467	0.9301	0.8703	-	0.9288	0.7471	-	0.9311	-	0.866	0.7976	0.9149	0.5106	-
completed	MANet	ResNet-50	focal+dice	0.9561	0.7133	0.8817	-	0.9115	0.6915	-	0.903	-	0.8808	0.8436	0.9347	0.8936	-
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.8025	0	0.8812	-	0.9131	0.5	-	0.8442	-	0.8116	0.7333	0.9041	0	-
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.9467	0.6783	0.8294	-	0.8701	0.6213	-	0.9147	-	0.9227	0.7654	0.9154	0	-
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.9216	0	0.8222	-	0.9144	0.5789	-	0.8795	-	0.8548	0.7759	0.9236	0	-
completed	DC-Swin	Swin-Small	ce+dice	0.6019	0	0.7729	-	0.8827	0.2851	-	0.7262	-	0.8328	0.7861	0.8852	0	-
completed	A2FPN	ResNet-18	ce+dice	0.6897	0	0.7552	-	0.8895	0.4708	-	0.7871	-	0.6587	0.5646	0.8499	0	-
completed	A2FPN	ResNet-18	weighted_ce+dice	0.7524	0.4685	0.7842	-	0.852	0.5556	-	0.8543	-	0.7668	0.468	0.8182	0.4894	-
completed	A2FPN	ResNet-18	focal+dice	0.8213	0	0.7625	-	0.8136	0.5921	-	0.8908	-	0.8408	0.6081	0.901	0	-
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.605	0.2517	0.89	-	0.904	0.5409	-	0.2533	-	0.7864	0.8112	0.8987	0	-
completed	FarSeg++	MiT-B2	ce (native)	0.5862	0.3357	0.7991	-	0.7771	0.2325	-	0.5262	-	0.6292	0.6632	0.8482	0	-
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.6489	0.0699	0.7695	-	0.7922	0.6447	-	0.5901	-	0.7961	0.6939	0.9109	0	-
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.6458	0.1888	0.81	-	0.8542	0.6725	-	0.4742	-	0.6902	0.6962	0.8787	0	-
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.8245	0	0.8056	-	0.8237	0.5731	-	0.8702	-	0.7436	0.7883	0.876	0	-
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.8746	0.4755	0.8345	-	0.846	0.6257	-	0.8341	-	0.8097	0.7147	0.8637	0.4149	-
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.8589	0	0.9066	-	0.8734	0.5599	-	0.9076	-	0.8589	0.6089	0.8842	0	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0	0	0.9185	-	0.9005	0	-	0	-	0.8162	0.8134	0.9278	0	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.9248	0	0.8515	-	0.9134	0.6915	-	0.7766	-	0.9386	0.8606	0.963	0	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0	0	0.9476	-	0.9232	0.3348	-	0.3167	-	0.8389	0.8085	0.934	0	-
completed	PyramidMamba	Swin-Base	ce+dice	0.8934	0.4755	0.8141	-	0.859	0.5906	-	0.8148	-	0.8028	0.8136	0.9013	0	-
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.9248	0.6014	0.8369	-	0.8844	0.769	-	0.8685	-	0.8023	0.8049	0.8933	0.3723	-
completed	PyramidMamba	Swin-Base	focal+dice	0.9091	0.4965	0.8059	-	0.8261	0.6711	-	0.8438	-	0.8416	0.752	0.9484	0	-
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.5956	0.0559	0.7633	-	0.7476	0.2061	-	0.3562	-	0.743	0.7058	0.8772	0	-
completed	MF-Mamba	HRNet-W18	ce+dice	0.7273	0	0.7922	-	0.7588	0.5965	-	0.8198	-	0.8016	0.6994	0.9182	0	-
completed	MCPNet	ResNet-50	ce+dice	0.6113	0	0.7916	-	0.7863	0.5453	-	0.6384	-	0.6521	0.6452	0.8947	0	-
completed	MCPNet	ResNet-50	weighted_ce+dice	0.5862	0.5804	0.7728	-	0.7506	0.5365	-	0.6287	-	0.7503	0.7443	0.8341	0	-
completed	MCPNet	ResNet-50	focal+dice	0.6771	0	0.8243	-	0.8487	0.5205	-	0.5825	-	0.7165	0.688	0.8465	0.0106	-

Methods related to vision foundation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.6865	0.2867	0.9189	-	0.9585	0.4664	-	0.6606	-	0.9328	0.96	0.9539	0	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.6708	0.5315	0.9004	-	0.9443	0.6462	-	0.8055	-	0.9615	0.9401	0.9482	0.0213	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.627	0.2867	0.9137	-	0.9473	0.3918	-	0.5443	-	0.9342	0.9527	0.9622	0	-
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.5256	0	0.692	-	0.8414	0.5044	-	0.5263	-	0.6798	0.6253	0.8097	0	-
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.6614	0	0.748	-	0.7663	0.3816	-	0.53	-	0.7055	0.6713	0.8477	0	-
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.4828	0	0.8219	-	0.7975	0.2822	-	0.7031	-	0.5805	0.6039	0.7995	0	-
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.6364	0	0.8008	-	0.8451	0.6784	-	0.7047	-	0.6804	0.6922	0.8956	0	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.7429	0.2657	0.9516	-	0.8607	0.1871	-	0.761	-	0.9362	0.888	0.9611	0	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.8715	0.6923	0.9483	-	0.8569	0.367	-	0.8744	-	0.9674	0.84	0.9589	0.1383	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.7962	0.3217	0.958	-	0.8732	0.2354	-	0.7396	-	0.9611	0.8887	0.9684	0	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.6677	0	0.7863	-	0.7587	0.2939	-	0.693	-	0.6469	0.711	0.8362	0	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.6426	0.7832	0.8553	-	0.8189	0.6023	-	0.6968	-	0.6978	0.5801	0.8499	0	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.7304	0	0.8626	-	0.7851	0.4985	-	0.7333	-	0.7674	0.535	0.858	0	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.7524	0	0.8379	-	0.8443	0.6067	-	0.8694	-	0.7665	0.7069	0.8961	0	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.7962	0.8462	0.8273	-	0.8847	0.6608	-	0.8492	-	0.8435	0.6133	0.8567	0.4362	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.8025	0.1818	0.9151	-	0.8675	0.5453	-	0.7812	-	0.7501	0.6202	0.8901	0	-
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.6897	0	0.7552	-	0.8895	0.4708	-	0.7871	-	0.6587	0.5646	0.8499	0	-
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.8213	0	0.7625	-	0.8136	0.5921	-	0.8908	-	0.8408	0.6081	0.901	0	-
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.7524	0.4685	0.7842	-	0.8521	0.5556	-	0.8543	-	0.7668	0.468	0.8182	0.4894	-
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.605	0	0.7599	-	0.8284	0.3991	-	0.743	-	0.7294	0.5424	0.7921	0	-
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.6364	0	0.7913	-	0.7884	0.4401	-	0.6888	-	0.7237	0.6612	0.8445	0	-
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.9231	0	0.8827	-	0.9247	0.677	-	0.8979	-	0.8471	0.8181	0.946	0	-
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.9561	0.7133	0.8817	-	0.9115	0.6915	-	0.903	-	0.8808	0.8436	0.9347	0.8936	-
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.9467	0.9301	0.8703	-	0.9288	0.7471	-	0.9311	-	0.866	0.7976	0.9149	0.5106	-
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.8025	0	0.8812	-	0.9131	0.5	-	0.8442	-	0.8116	0.7333	0.9041	0	-
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.9216	0	0.8222	-	0.9145	0.5789	-	0.8795	-	0.8548	0.7759	0.9236	0	-
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.9467	0.6783	0.8294	-	0.8701	0.6213	-	0.9147	-	0.9227	0.7654	0.9154	0	-

Test per-class f1

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/test_multilabel_per_class_f1.csv

General segmentation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.5939	0.4916	0.8251	-	0.8661	0.4102	-	0.6344	-	0.7037	0.6014	0.7709	0.2374	-
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.5728	0.4542	0.7956	-	0.8732	0.3479	-	0.5901	-	0.7149	0.5873	0.7669	0.3125	-
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.5936	0.4536	0.8083	-	0.8651	0.4324	-	0.6431	-	0.7069	0.5856	0.7795	0.0862	-
completed	DeepLabV3+	ResNet-50	ce+dice	0.6523	-	0.8194	-	0.8725	0.4295	-	0.6443	-	0.7047	0.5917	0.7518	-	-
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.6729	0.4865	0.8168	-	0.8486	0.2468	-	0.6664	-	0.7241	0.5609	0.7522	-	-
completed	DeepLabV3+	ResNet-50	focal+dice	0.4694	0.4967	0.8352	-	0.869	0.2771	-	0.6573	-	0.7059	0.5817	0.739	-	-
completed	UPerNet	Swin-Tiny	ce+dice	0.5324	0.3091	0.8082	-	0.8656	0.3937	-	0.5697	-	0.7051	0.5376	0.7639	-	-
completed	OCRNet	HRNet-W48	ce+dice	0.5709	-	0.6941	-	0.8088	0.2351	-	0.6486	-	0.7148	0.5756	0.7125	-	-
completed	SegFormer	MiT-B2	ce+dice	0.3756	0.3594	0.7784	-	0.818	0.3902	-	0.6545	-	0.7001	0.5718	0.7311	0.2180	-
completed	SegFormer	MiT-B2	weighted_ce+dice	0.6226	0.5798	0.8372	-	0.8914	0.5355	-	0.5991	-	0.7573	0.4656	0.7842	-	-
completed	SegFormer	MiT-B2	focal+dice	0.296	0.4968	0.81	-	0.7926	0.4639	-	0.6335	-	0.7095	0.6035	0.7159	0.2131	-
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.6077	0.4348	0.8194	-	0.7269	0.4487	-	0.681	-	0.7	0.552	0.7151	-	-
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.5026	0.2286	0.7631	-	0.8023	0.1934	-	0.5572	-	0.7192	0.5817	0.7245	-	-
completed	Afformer	AFFormer-Base	ce+dice	0.4366	0.3792	0.7892	-	0.7787	0.3474	-	0.5931	-	0.7267	0.5951	0.7155	0.1353	-
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.5413	0.3279	0.8307	-	0.8786	0.4021	-	0.5579	-	0.7151	0.5642	0.7987	-	-
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.4946	0.4667	0.8248	-	0.8706	0.3823	-	0.6608	-	0.7165	0.5913	0.7861	0.3194	-
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.5509	0.4940	0.8022	-	0.8616	0.4443	-	0.6182	-	0.7155	0.561	0.7682	0.2400	-
completed	SeaFormer	SeaFormer-Base	ce+dice	0.4777	0.1818	0.8078	-	0.7835	0.5055	-	0.5808	-	0.7362	0.5645	0.7089	-	-
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.4978	0.3277	0.7787	-	0.7721	0.3091	-	0.5336	-	0.7085	0.5563	0.7049	0.0583	-
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.5202	0.2412	0.8336	-	0.7024	0.5593	-	0.6283	-	0.7163	0.5592	0.6949	-	-

Remote-sensing-specific methods

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	FarSeg	ResNet-50	ce (native)	0.5925	-	0.8051	-	0.8933	0.3569	-	0.5930	-	0.723	0.5989	0.7932	-	-
completed	FarSeg	ResNet-50	ce+dice	0.5564	-	0.7547	-	0.8888	0.3416	-	0.5353	-	0.7255	0.6158	0.7808	-	-
completed	FarSeg	ResNet-50	weighted_ce+dice	0.2652	0.3295	0.7696	-	0.8666	0.3615	-	0.5254	-	0.6504	0.5685	0.7568	0.3617	-
completed	FarSeg	ResNet-50	focal+dice	0.5394	-	0.847	-	0.8959	0.3319	-	0.7288	-	0.7281	0.6107	0.7758	-	-
completed	BANet	ResT-Lite	ce+dice	0.4342	-	0.8144	-	0.805	0.3899	-	0.5463	-	0.6995	0.5767	0.7248	-	-
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.5893	-	0.8138	-	0.8642	0.3391	-	0.6776	-	0.7153	0.5203	0.7547	-	-
completed	MANet	ResNet-50	ce+dice	0.5924	-	0.7979	-	0.8628	0.3572	-	0.5627	-	0.7265	0.5965	0.7402	-	-
completed	MANet	ResNet-50	weighted_ce+dice	0.5355	0.2603	0.8073	-	0.8545	0.2753	-	0.4907	-	0.7216	0.5878	0.7526	0.2494	-
completed	MANet	ResNet-50	focal+dice	0.3888	0.2873	0.826	-	0.8623	0.3103	-	0.5072	-	0.7063	0.5914	0.7202	0.2445	-
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.6641	-	0.831	-	0.8755	0.5662	-	0.6208	-	0.749	0.6188	0.7588	-	-
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.3654	0.4491	0.8347	-	0.8727	0.2961	-	0.4452	-	0.7184	0.5952	0.7271	-	-
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.5568	-	0.8276	-	0.8852	0.4554	-	0.5489	-	0.7282	0.6194	0.7417	-	-
completed	DC-Swin	Swin-Small	ce+dice	0.4747	-	0.8144	-	0.8602	0.2968	-	0.4934	-	0.7178	0.6054	0.7411	-	-
completed	A2FPN	ResNet-18	ce+dice	0.6086	-	0.8133	-	0.8862	0.3868	-	0.6286	-	0.6758	0.544	0.7993	-	-
completed	A2FPN	ResNet-18	weighted_ce+dice	0.6170	0.4527	0.8267	-	0.8706	0.4241	-	0.5826	-	0.7208	0.4983	0.7593	0.3119	-
completed	A2FPN	ResNet-18	focal+dice	0.6231	-	0.8057	-	0.8505	0.3343	-	0.5007	-	0.7184	0.5485	0.7344	-	-
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.6328	0.3789	0.8479	-	0.8577	0.5305	-	0.3573	-	0.7033	0.6089	0.7583	-	-
completed	FarSeg++	MiT-B2	ce (native)	0.4934	0.3983	0.8239	-	0.8195	0.2602	-	0.5459	-	0.6741	0.596	0.728	-	-
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.5948	0.1307	0.7916	-	0.8419	0.5326	-	0.6411	-	0.734	0.5719	0.7338	-	-
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.5722	0.3140	0.8037	-	0.8512	0.3907	-	0.6091	-	0.6987	0.5876	0.7593	-	-
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.5998	-	0.8231	-	0.8462	0.4891	-	0.6484	-	0.711	0.5967	0.7305	-	-
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.5586	0.5191	0.8322	-	0.8496	0.3887	-	0.6104	-	0.7078	0.5897	0.7266	0.4906	-
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.5963	-	0.8311	-	0.8611	0.5933	-	0.5763	-	0.7365	0.5527	0.7685	-	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	-	-	0.8569	-	0.8671	-	-	-	-	0.734	0.5869	0.7202	-	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.1652	-	0.8418	-	0.8342	0.2752	-	0.4392	-	0.643	0.5909	0.6289	-	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	-	-	0.8522	-	0.8708	0.3955	-	0.4499	-	0.735	0.5832	0.724	-	-
completed	PyramidMamba	Swin-Base	ce+dice	0.6051	0.5667	0.8208	-	0.8591	0.3988	-	0.7015	-	0.7309	0.5866	0.7424	-	-
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.4966	0.6014	0.8046	-	0.8476	0.2091	-	0.6757	-	0.7289	0.6093	0.7507	0.4575	-
completed	PyramidMamba	Swin-Base	focal+dice	0.5441	0.3413	0.7982	-	0.8547	0.3971	-	0.6041	-	0.7315	0.5896	0.7038	-	-
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.3050	0.1039	0.7984	-	0.7759	0.2000	-	0.4497	-	0.685	0.585	0.7249	-	-
completed	MF-Mamba	HRNet-W18	ce+dice	0.5027	-	0.8152	-	0.8093	0.2650	-	0.5562	-	0.7071	0.5781	0.7226	-	-
completed	MCPNet	ResNet-50	ce+dice	0.4235	-	0.8162	-	0.8111	0.3775	-	0.6619	-	0.69	0.5767	0.7291	-	-
completed	MCPNet	ResNet-50	weighted_ce+dice	0.4634	0.3242	0.8147	-	0.7986	0.4993	-	0.5915	-	0.7239	0.6102	0.718	-	-
completed	MCPNet	ResNet-50	focal+dice	0.2488	-	0.8287	-	0.854	0.4191	-	0.5460	-	0.7167	0.591	0.7465	0.0206	-

Methods related to vision foundation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.265	0.2742	0.8317	-	0.8034	0.1554	-	0.4401	-	0.6075	0.5295	0.6816	-	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.1885	0.1317	0.837	-	0.8043	0.09	-	0.3779	-	0.5638	0.5392	0.6825	0.0114	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.3891	0.2571	0.842	-	0.823	0.1538	-	0.4547	-	0.6074	0.535	0.6704	-	-
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.4706	-	0.7746	-	0.8683	0.377	-	0.6583	-	0.6995	0.5612	0.7438	-	-
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.596	-	0.7991	-	0.8265	0.3537	-	0.6539	-	0.7021	0.5735	0.7524	-	-
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.5283	-	0.8385	-	0.8197	0.2546	-	0.5327	-	0.6512	0.5877	0.7252	-	-
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.4683	-	0.8209	-	0.8461	0.2873	-	0.6799	-	0.7035	0.6103	0.7374	-	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.2931	0.2764	0.8286	-	0.7907	0.1616	-	0.3809	-	0.631	0.5655	0.6772	-	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.0811	0.0733	0.8244	-	0.7743	0.0833	-	0.3363	-	0.5908	0.5555	0.6794	0.0327	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.1646	0.2939	0.8265	-	0.7839	0.1003	-	0.4316	-	0.5924	0.5522	0.6685	-	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.4919	-	0.8156	-	0.8086	0.3076	-	0.4769	-	0.6901	0.5747	0.7276	-	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.2929	0.1856	0.8314	-	0.8328	0.2342	-	0.4182	-	0.6931	0.5589	0.7348	-	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.424	-	0.8426	-	0.8277	0.3034	-	0.5269	-	0.7127	0.5422	0.7347	-	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.4356	-	0.8117	-	0.8568	0.3332	-	0.4318	-	0.7229	0.5892	0.7556	-	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.5163	0.2919	0.833	-	0.8503	0.3877	-	0.5205	-	0.7216	0.5643	0.7508	0.2297	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.3829	0.2938	0.8451	-	0.8635	0.4645	-	0.5876	-	0.7322	0.5965	0.7666	-	-
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.6086	-	0.8133	-	0.8861	0.3868	-	0.6286	-	0.6758	0.544	0.7993	-	-
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.6231	-	0.8057	-	0.8503	0.3343	-	0.5007	-	0.7184	0.5485	0.7344	-	-
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.617	0.4527	0.8267	-	0.8704	0.4241	-	0.5826	-	0.7208	0.4983	0.7593	0.3119	-
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.5893	-	0.8138	-	0.8641	0.3391	-	0.6776	-	0.7153	0.5203	0.7547	-	-
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.4342	-	0.8144	-	0.805	0.3899	-	0.5463	-	0.6995	0.5767	0.7248	-	-
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.538	-	0.8104	-	0.8524	0.3014	-	0.564	-	0.7374	0.5885	0.7521	-	-
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.3888	0.2873	0.826	-	0.8623	0.3103	-	0.5072	-	0.7063	0.5914	0.7202	0.2445	-
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.5355	0.2603	0.8073	-	0.8544	0.2753	-	0.4907	-	0.7216	0.5878	0.7525	0.2494	-
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.6641	-	0.831	-	0.8753	0.5662	-	0.6208	-	0.749	0.6188	0.7588	-	-
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.5568	-	0.8276	-	0.885	0.4554	-	0.5489	-	0.7282	0.6194	0.7417	-	-
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.3654	0.4491	0.8347	-	0.8724	0.2961	-	0.4452	-	0.7184	0.5952	0.7271	-	-

Test per-class ap

• Source: /deac/csc/yangGrp/cuij/GoldMDD/experiments/diagnostics/test_multilabel_per_class_ap.csv

General segmentation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	DeepLabV3+	ConvNeXt-Tiny	ce+dice	0.6052	0.4658	0.8194	-	0.9275	0.5027	-	0.748	-	0.7153	0.4888	0.745	0.1207	-
completed	DeepLabV3+	ConvNeXt-Tiny	weighted_ce+dice	0.6532	0.376	0.8261	-	0.9182	0.357	-	0.7399	-	0.7008	0.4437	0.6979	0.224	-
completed	DeepLabV3+	ConvNeXt-Tiny	focal+dice	0.6245	0.4539	0.8287	-	0.9246	0.4409	-	0.7084	-	0.7101	0.4718	0.7239	0.0396	-
completed	DeepLabV3+	ResNet-50	ce+dice	0.7532	0.0189	0.8289	-	0.9254	0.4024	-	0.7955	-	0.7086	0.4885	0.6508	0.01	-
completed	DeepLabV3+	ResNet-50	weighted_ce+dice	0.717	0.4056	0.8288	-	0.9044	0.1567	-	0.7362	-	0.7037	0.4822	0.6393	0.01	-
completed	DeepLabV3+	ResNet-50	focal+dice	0.6571	0.4209	0.85	-	0.9307	0.2984	-	0.741	-	0.7182	0.4814	0.6882	0.01	-
completed	UPerNet	Swin-Tiny	ce+dice	0.4256	0.191	0.8053	-	0.9051	0.3399	-	0.5622	-	0.6561	0.4232	0.6929	0.01	-
completed	OCRNet	HRNet-W48	ce+dice	0.4672	0.0189	0.7802	-	0.8503	0.2013	-	0.6585	-	0.6977	0.4206	0.528	0.01	-
completed	SegFormer	MiT-B2	ce+dice	0.4826	0.3052	0.7937	-	0.8617	0.2905	-	0.7258	-	0.702	0.4548	0.5543	0.0638	-
completed	SegFormer	MiT-B2	weighted_ce+dice	0.7568	0.5136	0.82	-	0.9244	0.5274	-	0.7699	-	0.7207	0.4213	0.7023	0.01	-
completed	SegFormer	MiT-B2	focal+dice	0.6103	0.4562	0.8228	-	0.8389	0.4665	-	0.7213	-	0.7217	0.4734	0.5231	0.0571	-
completed	Mask2Former	ResNet-50	set_matching_ce+mask+dice	0.5012	0.2957	0.8039	-	0.7864	0.2685	-	0.6755	-	0.6778	0.4819	0.5934	0.01	-
completed	SegNeXt	MSCAN-Tiny	ce+dice	0.5003	0.1381	0.8036	-	0.851	0.1061	-	0.5822	-	0.6999	0.4166	0.5746	0.01	-
completed	Afformer	AFFormer-Base	ce+dice	0.5499	0.2323	0.8121	-	0.833	0.3415	-	0.6212	-	0.7235	0.4558	0.5381	0.0461	-
completed	EfficientViT-Seg	EfficientViT-B2	ce+dice	0.5544	0.2311	0.8274	-	0.9165	0.5232	-	0.657	-	0.6978	0.4614	0.746	0.01	-
completed	EfficientViT-Seg	EfficientViT-B2	weighted_ce+dice	0.5431	0.3716	0.8295	-	0.916	0.4473	-	0.7063	-	0.6907	0.4719	0.72	0.2073	-
completed	EfficientViT-Seg	EfficientViT-B2	focal+dice	0.4667	0.4423	0.8391	-	0.9075	0.4824	-	0.7242	-	0.6992	0.4349	0.6618	0.1182	-
completed	SeaFormer	SeaFormer-Base	ce+dice	0.4831	0.11	0.8054	-	0.83	0.4067	-	0.6018	-	0.7317	0.4431	0.5206	0.01	-
completed	CGRSeg	EfficientFormerV2-B	ce+dice	0.5475	0.2301	0.8074	-	0.8201	0.2376	-	0.557	-	0.6735	0.4046	0.5192	0.0231	-
completed	PEM	ResNet-50	set_matching_ce+mask+dice	0.3711	0.1412	0.8212	-	0.7481	0.4424	-	0.6294	-	0.6638	0.5076	0.5334	0.01	-

Remote-sensing-specific methods

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	FarSeg	ResNet-50	ce (native)	0.5609	0.0189	0.8159	-	0.9365	0.2624	-	0.6678	-	0.704	0.485	0.8019	0.01	-
completed	FarSeg	ResNet-50	ce+dice	0.6463	0.0189	0.8007	-	0.9154	0.4912	-	0.7195	-	0.689	0.4522	0.8192	0.01	-
completed	FarSeg	ResNet-50	weighted_ce+dice	0.5255	0.2317	0.8135	-	0.9243	0.4754	-	0.7647	-	0.6114	0.4361	0.6766	0.3031	-
completed	FarSeg	ResNet-50	focal+dice	0.6595	0.0189	0.8512	-	0.9454	0.3198	-	0.7362	-	0.7057	0.4941	0.8416	0.01	-
completed	BANet	ResT-Lite	ce+dice	0.4034	0.0185	0.8114	-	0.8521	0.2936	-	0.5988	-	0.671	0.4584	0.547	0.01	-
completed	ABCNet	ResNet-18	ce+dice+aux_ce	0.45	0.0189	0.804	-	0.8934	0.2952	-	0.7071	-	0.6422	0.4369	0.6587	0.01	-
completed	MANet	ResNet-50	ce+dice	0.8302	0.0181	0.8444	-	0.9393	0.4726	-	0.7492	-	0.7051	0.4388	0.6855	0.01	-
completed	MANet	ResNet-50	weighted_ce+dice	0.8666	0.6568	0.844	-	0.9322	0.4933	-	0.7667	-	0.6882	0.4221	0.6779	0.2113	-
completed	MANet	ResNet-50	focal+dice	0.8194	0.5011	0.8504	-	0.9284	0.5012	-	0.7777	-	0.7159	0.4493	0.6484	0.4515	-
completed	UNetFormer	ResNet-18	ce+dice+aux_ce	0.7397	0.0189	0.8217	-	0.9336	0.493	-	0.7374	-	0.7264	0.4957	0.672	0.01	-
completed	UNetFormer	ResNet-18	weighted_ce+dice+aux_ce	0.7636	0.4324	0.8399	-	0.9235	0.3679	-	0.7284	-	0.684	0.454	0.6225	0.01	-
completed	UNetFormer	ResNet-18	focal+dice+aux_focal	0.7687	0.0189	0.8402	-	0.9316	0.4355	-	0.6966	-	0.6914	0.5004	0.7197	0.01	-
completed	DC-Swin	Swin-Small	ce+dice	0.428	0.0189	0.8201	-	0.9109	0.2035	-	0.6009	-	0.6889	0.4542	0.5905	0.01	-
completed	A2FPN	ResNet-18	ce+dice	0.5576	0.0189	0.8095	-	0.9243	0.2816	-	0.7047	-	0.6389	0.4708	0.8223	0.01	-
completed	A2FPN	ResNet-18	weighted_ce+dice	0.6513	0.3785	0.8072	-	0.9099	0.3942	-	0.7274	-	0.6695	0.453	0.6719	0.3189	-
completed	A2FPN	ResNet-18	focal+dice	0.7207	0.0189	0.8215	-	0.8943	0.3064	-	0.7211	-	0.6777	0.4353	0.6133	0.01	-
completed	LoGCAN	ResNet-50	ce+aux_ce (native)	0.5255	0.2614	0.8496	-	0.91	0.4964	-	0.3804	-	0.6881	0.5363	0.6775	0.01	-
completed	FarSeg++	MiT-B2	ce (native)	0.4468	0.3042	0.8162	-	0.862	0.1655	-	0.54	-	0.6702	0.4827	0.5669	0.0099	-
completed	SACANet	HRNet-W32	ce+aux_ce (native)	0.5265	0.0935	0.8374	-	0.8724	0.5306	-	0.6368	-	0.7136	0.4366	0.6241	0.01	-
completed	DOCNet	HRNet-W32	ce+aux_ce (native)	0.4204	0.2258	0.821	-	0.8827	0.3717	-	0.6047	-	0.6746	0.454	0.6803	0.01	-
completed	PPMambaSeg	swsl-ResNet-18	ce+dice	0.6582	0.0189	0.8313	-	0.897	0.4797	-	0.7839	-	0.6724	0.4708	0.58	0.01	-
completed	PPMambaSeg	swsl-ResNet-18	weighted_ce+dice	0.7491	0.405	0.8324	-	0.9055	0.4652	-	0.7539	-	0.6691	0.4581	0.6029	0.3785	-
completed	PPMambaSeg	swsl-ResNet-18	focal+dice	0.7525	0.0189	0.8294	-	0.9193	0.5382	-	0.7644	-	0.7042	0.4355	0.652	0.01	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	ce+dice	0.015	0.0189	0.835	-	0.9211	0.0453	-	0.2438	-	0.7098	0.5734	0.6743	0.01	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	weighted_ce+dice	0.2461	0.0189	0.8381	-	0.9119	0.5113	-	0.4579	-	0.6828	0.4594	0.5796	0.01	-
completed	RS3Mamba	ResNet-18 + VMamba-Tiny	focal+dice	0.015	0.0189	0.8337	-	0.9247	0.3136	-	0.4751	-	0.7018	0.5669	0.6947	0.01	-
completed	PyramidMamba	Swin-Base	ce+dice	0.8151	0.4789	0.8273	-	0.9148	0.4122	-	0.7793	-	0.7308	0.4671	0.6143	0.01	-
completed	PyramidMamba	Swin-Base	weighted_ce+dice	0.8546	0.5716	0.8331	-	0.9176	0.5685	-	0.8074	-	0.7448	0.4954	0.6073	0.3488	-
completed	PyramidMamba	Swin-Base	focal+dice	0.8613	0.4011	0.825	-	0.9023	0.4437	-	0.7756	-	0.7288	0.4502	0.5961	0.01	-
completed	LoGCAN++	RepViT-M2.3	ce+aux_ce (native)	0.2195	0.0827	0.7894	-	0.817	0.173	-	0.4654	-	0.6348	0.4529	0.5453	0.01	-
completed	MF-Mamba	HRNet-W18	ce+dice	0.4628	0.0189	0.8232	-	0.8577	0.3328	-	0.6314	-	0.6567	0.4514	0.5622	0.01	-
completed	MCPNet	ResNet-50	ce+dice	0.2473	0.0186	0.8107	-	0.8566	0.2788	-	0.6732	-	0.6966	0.4893	0.568	0.0099	-
completed	MCPNet	ResNet-50	weighted_ce+dice	0.2923	0.3333	0.8132	-	0.8431	0.4577	-	0.6093	-	0.7335	0.4871	0.5613	0.01	-
completed	MCPNet	ResNet-50	focal+dice	0.2555	0.0189	0.8211	-	0.8938	0.4785	-	0.5638	-	0.7134	0.4863	0.6041	0.0136	-

Methods related to vision foundation models

status	model	backbone	loss	Building	Mining raft	Primary Forest	Heavy machinery	Water bodies	Agricultural crop	Compact mounds	Gravel mounds	Grass	Type1 regen	Type2 regen	Bare ground	Sluice	Vehicles
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	ce+dice	0.4167	0.1321	0.8749	-	0.9146	0.1273	-	0.5248	-	0.7358	0.4594	0.5535	0.01	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	weighted_ce+dice	0.3655	0.1955	0.8802	-	0.8954	0.1129	-	0.4819	-	0.721	0.439	0.5242	0.0095	-
completed	HQ-SAM	ViT-B + HQ decoder (full finetune, msfpn)	focal+dice	0.4746	0.181	0.8791	-	0.9106	0.1158	-	0.4782	-	0.7145	0.4888	0.5688	0.01	-
completed	SAM_RS	ABCNet + SAM priors	seg+bdy+obj (native)	0.2258	0.0189	0.798	-	0.9012	0.3196	-	0.635	-	0.6556	0.4541	0.6473	0.01	-
completed	SAM_RS	CMTFNet + SAM priors	seg+bdy+obj (native)	0.4057	0.0189	0.8107	-	0.8679	0.1417	-	0.6313	-	0.668	0.4534	0.6074	0.01	-
completed	SAM_RS	FTUNetFormer + SAM priors	seg+bdy+obj (native)	0.3986	0.0189	0.8033	-	0.8646	0.171	-	0.5149	-	0.6742	0.5325	0.5399	0.01	-
completed	SAM_RS	UNetFormer + SAM priors	seg+bdy+obj (native)	0.4161	0.0189	0.8251	-	0.8991	0.269	-	0.6813	-	0.7122	0.4914	0.6211	0.01	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	ce+dice	0.4801	0.1719	0.8524	-	0.8667	0.0801	-	0.5118	-	0.7099	0.4239	0.5145	0.01	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	weighted_ce+dice	0.1094	0.1042	0.8711	-	0.8678	0.061	-	0.4988	-	0.671	0.4103	0.5014	0.0118	-
completed	SAM2.1	Hiera-B+ (frozen backbone, msfpn)	focal+dice	0.3368	0.2089	0.8769	-	0.8726	0.0737	-	0.5452	-	0.6877	0.4242	0.5045	0.01	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	ce+dice	0.4621	0.0189	0.8062	-	0.848	0.2161	-	0.542	-	0.693	0.4542	0.5449	0.01	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	weighted_ce+dice	0.2097	0.3964	0.8205	-	0.8782	0.3169	-	0.4935	-	0.6721	0.4716	0.581	0.01	-
completed	SAM2.1	Hiera-B+ (full finetune, msfpn)	focal+dice	0.383	0.0189	0.8162	-	0.8574	0.2637	-	0.5548	-	0.6786	0.4799	0.5779	0.01	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	ce+dice	0.5349	0.0189	0.8133	-	0.904	0.3774	-	0.5414	-	0.7021	0.4758	0.668	0.01	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	weighted_ce+dice	0.6478	0.453	0.8247	-	0.906	0.4637	-	0.6663	-	0.6742	0.4601	0.6346	0.1382	-
completed	RSAM-Seg	SAM-ViT-B (frozen encoder)	focal+dice	0.4918	0.1961	0.8341	-	0.9069	0.3793	-	0.6757	-	0.7176	0.5535	0.6892	0.01	-
completed	SESSRS	A2FPN (ce+dice)	t1/t2 search + postprocess	0.5604	0.0189	0.8095	-	0.9243	0.2812	-	0.7034	-	0.639	0.4708	0.8223	0.01	-
completed	SESSRS	A2FPN (focal)	t1/t2 search + postprocess	0.7206	0.0189	0.8215	-	0.8942	0.3063	-	0.7198	-	0.6778	0.4354	0.6132	0.01	-
completed	SESSRS	A2FPN (weighted)	t1/t2 search + postprocess	0.6534	0.3786	0.8071	-	0.9099	0.3939	-	0.726	-	0.6695	0.4531	0.6719	0.3191	-
completed	SESSRS	ABCNet (ce+dice+aux)	t1/t2 search + postprocess	0.4507	0.0189	0.804	-	0.8935	0.2952	-	0.7062	-	0.6423	0.4369	0.6587	0.01	-
completed	SESSRS	BANet (ce+dice)	t1/t2 search + postprocess	0.4024	0.0185	0.8114	-	0.852	0.2933	-	0.6	-	0.671	0.4585	0.547	0.01	-
completed	SESSRS	MANet (ce+dice)	t1/t2 search + postprocess	0.7823	0.0194	0.8482	-	0.9273	0.4235	-	0.7551	-	0.7357	0.4317	0.684	0.0102	-
completed	SESSRS	MANet (focal)	t1/t2 search + postprocess	0.8213	0.5013	0.8504	-	0.9285	0.5011	-	0.7783	-	0.7158	0.4493	0.6483	0.451	-
completed	SESSRS	MANet (weighted)	t1/t2 search + postprocess	0.8665	0.6566	0.844	-	0.9322	0.4919	-	0.7666	-	0.6882	0.4221	0.6778	0.2113	-
completed	SESSRS	UNetFormer (ce+dice)	t1/t2 search + postprocess	0.7391	0.0189	0.8217	-	0.9338	0.4927	-	0.7362	-	0.7263	0.4957	0.6716	0.01	-
completed	SESSRS	UNetFormer (focal)	t1/t2 search + postprocess	0.7689	0.0189	0.8402	-	0.9317	0.4354	-	0.6956	-	0.6916	0.5005	0.7197	0.01	-
completed	SESSRS	UNetFormer (weighted)	t1/t2 search + postprocess	0.7647	0.4332	0.8399	-	0.9235	0.3672	-	0.729	-	0.6839	0.4542	0.6225	0.01	-