Upload folder using huggingface_hub

.gitattributes

@@ -52,3 +52,153 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 10samples/sample_0007/overlays/overlay_accepted.png filter=lfs diff=lfs merge=lfs -text
 10samples/sample_0007/overlays/overlay_intended.png filter=lfs diff=lfs merge=lfs -text
 10samples/sample_0007/overlays/overlay_measured.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/crops/detect_refine_metal_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/crops/detect_refine_parked_dark_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/crops/detect_refine_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/crops/diversify_input_metal_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/crops/diversify_input_parked_dark_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/crops/diversify_input_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/references/ref_metal_barrier.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/references/ref_parked_dark_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000001/references/ref_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/crops/detect_refine_street_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/crops/detect_refine_twilight_sky.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/crops/detect_refine_vehicle_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/crops/diversify_input_double_yellow_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/crops/diversify_input_street_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/crops/diversify_input_twilight_sky.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_city_buildings.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_dark_car_left.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_dark_suv_driving.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_double_yellow_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_parked_suv_right.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_pedestrian_left.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_pedestrian_right.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_pink_scooter.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_storefront_sign.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_street_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_street_signs.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_street_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_twilight_sky.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_vehicle_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000002/references/ref_white_car_ahead.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/crops/detect_refine_black_sedan.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/crops/detect_refine_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/crops/diversify_input_black_sedan.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/crops/diversify_input_shopper.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/crops/diversify_input_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/references/ref_black_sedan.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/references/ref_shopper.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000003/references/ref_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/crops/detect_refine_dark_parked_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/crops/detect_refine_plain_delivery_truck.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/crops/detect_refine_street_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/crops/diversify_input_dark_parked_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/crops/diversify_input_plain_delivery_truck.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/crops/diversify_input_street_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/references/ref_dark_parked_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/references/ref_pedestrian_walker.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/references/ref_plain_delivery_truck.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/references/ref_red_traffic_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000004/references/ref_street_lines.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/crops/detect_refine_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/crops/detect_refine_street_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/crops/diversify_input_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/references/ref_man_in_suit.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/references/ref_person_yellow_top.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/references/ref_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/references/ref_street_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/references/ref_traffic_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000005/references/ref_young_girl.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/crops/detect_refine_blooming_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/crops/detect_refine_double_solid_white_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/crops/detect_refine_white_panel_van.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/crops/diversify_input_blooming_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/crops/diversify_input_double_solid_white_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/crops/diversify_input_white_panel_van.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/references/ref_blooming_trees.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/references/ref_curbside_trash_can.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/references/ref_double_solid_white_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/references/ref_emergency_vehicle.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/references/ref_green_street_sign.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/references/ref_museum_banner.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/references/ref_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000006/references/ref_white_panel_van.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000007/crops/detect_refine_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000007/crops/detect_refine_multi_story_building_left.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000007/crops/detect_refine_overhead_wires.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000007/crops/detect_refine_red_brick_building.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000007/crops/detect_refine_street_light_pole.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000007/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/crops/detect_refine_dashboard_reflection.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/crops/detect_refine_passenger.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/crops/diversify_input_dashboard_reflection.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/crops/diversify_input_passenger.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/crops/diversify_input_pedestrian_crossing.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/crops/diversify_input_shopper_waiting.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_businessman.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_dashboard_reflection.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_passenger.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_pedestrian_crossing.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_pedestrian_walking_away.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_protester.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_shopper_standing.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_shopper_waiting.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_street_lamp.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000008/references/ref_young_man.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/crops/detect_refine_overpass.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/crops/detect_refine_yellow_building.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/crops/detect_refine_yellow_lane_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/crops/diversify_input_overpass.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/crops/diversify_input_yellow_building.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/crops/diversify_input_yellow_lane_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/references/ref_overpass.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/references/ref_pedestrian_in_suit.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/references/ref_street_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/references/ref_yellow_building.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000009/references/ref_yellow_lane_line.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/crops/detect_refine_black_suv.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/crops/diversify_input_black_suv.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/crops/diversify_input_pedestrian_black_jacket.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_black_suv.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_distant_pedestrian.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_man_pink_shirt.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_pedestrian_backpack.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_pedestrian_black_jacket.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_pedestrian_light_blue_shirt.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_pedestrian_light_jacket.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_pedestrian_red_jacket.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000010/references/ref_pedestrian_striped_shirt.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/bbox_overlay.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/crops/detect_refine_ego_car_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/crops/diversify_input_ego_car_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/crops/diversify_input_windshield_mount.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/main_image.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_background_street_light.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_blonde_woman.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_ego_car_dashboard.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_metal_structure.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_oncoming_white_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_overhead_streetlights.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_silver_car.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_vintage_street_lamp.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_white_suv.png filter=lfs diff=lfs merge=lfs -text
+samples_v8/driving/BDD100K_CrowdHuman_samples/sample_000011/references/ref_windshield_mount.png filter=lfs diff=lfs merge=lfs -text

samples_v8/driving/BDD100K_CrowdHuman_samples/README.md

@@ -0,0 +1,16 @@
+# samples_v8
+
+Generated with `data_recipe_v8.md`: vocabulary-first planning, adaptive canvas selection, structured JSON compose prompts, no identity verification, no gate, SAM white-background reference postprocessing, and strict reference-completeness verification with regenerate-until-pass behavior.
+
+- chat model: `gcp/google/gemini-3.1-pro-preview`
+- image model: `gcp/google/gemini-3-pro-image-preview`
+- people references: `white_bg_full_body_front`
+- non-person references: `white_bg_encyclopedia_photo`
+- SAM postprocess: every generated reference is segmented with `sam_vit_b` and pasted onto pure `#ffffff` background
+- reference verify max attempts per subject: `10`
+- allowed canvases: `[{"aspect_ratio": "1:1", "size": [1024, 1024], "style": "photorealistic"}, {"aspect_ratio": "4:3", "size": [1152, 864], "style": "photorealistic"}, {"aspect_ratio": "3:4", "size": [864, 1152], "style": "photorealistic"}, {"aspect_ratio": "3:2", "size": [1248, 832], "style": "photorealistic"}, {"aspect_ratio": "2:3", "size": [832, 1248], "style": "photorealistic"}, {"aspect_ratio": "16:9", "size": [1280, 720], "style": "photorealistic"}, {"aspect_ratio": "9:16", "size": [720, 1280], "style": "photorealistic"}]`
+- scenario mode: `driving`
+- pools: `vocab_task_pool`, `plan_pool`, `scene_pool`, `detection_pool`, `reference_pool`
+- bbox overlay: `bbox_overlay.png` draws every planned subject bbox; a sample is rejected and regenerated if any planned subject is still missing after VLM detection retries
+- detection max attempts per subject: `3`
+- launch args: `{"compose_workers": 3, "detect_max_attempts": 3, "detect_workers": 3, "emit_workers": 4, "idle_sleep": 1.0, "image_inflight": 32, "image_interval": 0.05, "image_max_retries": 8, "max_retries": 3, "no_topup": false, "plan_workers": 6, "ref_verify_max_attempts": 10, "reference_workers": 6, "requeue_in_progress": true, "seed": 1781927993, "status_interval": 30.0, "subject_detect_workers": 24, "target_samples": 10}`

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000001.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000001",
+  "plan_path": "sample_000001/plan.json",
+  "task_path": "sample_000001/vocab_task.json",
+  "main_image": "sample_000001/main_image.png",
+  "detections": "sample_000001/detections.json",
+  "n_detected": 3,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000001",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000002.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000002",
+  "plan_path": "sample_000002/plan.json",
+  "task_path": "sample_000002/vocab_task.json",
+  "main_image": "sample_000002/main_image.png",
+  "detections": "sample_000002/detections.json",
+  "n_detected": 15,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000002",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000003.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000003",
+  "plan_path": "sample_000003/plan.json",
+  "task_path": "sample_000003/vocab_task.json",
+  "main_image": "sample_000003/main_image.png",
+  "detections": "sample_000003/detections.json",
+  "n_detected": 3,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000003",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000004.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000004",
+  "plan_path": "sample_000004/plan.json",
+  "task_path": "sample_000004/vocab_task.json",
+  "main_image": "sample_000004/main_image.png",
+  "detections": "sample_000004/detections.json",
+  "n_detected": 5,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000004",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000005.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000005",
+  "plan_path": "sample_000005/plan.json",
+  "task_path": "sample_000005/vocab_task.json",
+  "main_image": "sample_000005/main_image.png",
+  "detections": "sample_000005/detections.json",
+  "n_detected": 6,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000005",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000006.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000006",
+  "plan_path": "sample_000006/plan.json",
+  "task_path": "sample_000006/vocab_task.json",
+  "main_image": "sample_000006/main_image.png",
+  "detections": "sample_000006/detections.json",
+  "n_detected": 8,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000006",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000008.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000008",
+  "plan_path": "sample_000008/plan.json",
+  "task_path": "sample_000008/vocab_task.json",
+  "main_image": "sample_000008/main_image.png",
+  "detections": "sample_000008/detections.json",
+  "n_detected": 10,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000008",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000009.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000009",
+  "plan_path": "sample_000009/plan.json",
+  "task_path": "sample_000009/vocab_task.json",
+  "main_image": "sample_000009/main_image.png",
+  "detections": "sample_000009/detections.json",
+  "n_detected": 5,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000009",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000010.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000010",
+  "plan_path": "sample_000010/plan.json",
+  "task_path": "sample_000010/vocab_task.json",
+  "main_image": "sample_000010/main_image.png",
+  "detections": "sample_000010/detections.json",
+  "n_detected": 9,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000010",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/detection_pool/done/sample_000011.json

@@ -0,0 +1,16 @@
+{
+  "sample_id": "sample_000011",
+  "plan_path": "sample_000011/plan.json",
+  "task_path": "sample_000011/vocab_task.json",
+  "main_image": "sample_000011/main_image.png",
+  "detections": "sample_000011/detections.json",
+  "n_detected": 10,
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000011",
+  "pool": "detection_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000001.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000001",
+  "row": "sample_000001/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000002.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000002",
+  "row": "sample_000002/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000003.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000003",
+  "row": "sample_000003/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000004.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000004",
+  "row": "sample_000004/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000005.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000005",
+  "row": "sample_000005/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000006.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000006",
+  "row": "sample_000006/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000008.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000008",
+  "row": "sample_000008/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000009.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000009",
+  "row": "sample_000009/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000010.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000010",
+  "row": "sample_000010/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/emit_pool/done/sample_000011.json

@@ -0,0 +1,4 @@
+{
+  "sample_id": "sample_000011",
+  "row": "sample_000011/row.json"
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000001.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000001",
+  "plan_path": "sample_000001/plan.json",
+  "task_path": "sample_000001/vocab_task.json",
+  "prompt_hash": "ee63c678fc09a67bb20b6d08e8ef2b19732be3312cf3b823e6d07e1c5dd44f63",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000001",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000002.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000002",
+  "plan_path": "sample_000002/plan.json",
+  "task_path": "sample_000002/vocab_task.json",
+  "prompt_hash": "c3e70d0e58500cbcc95ef2a96d5d6793951917c615216803cc4c45dcc2a3a379",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000002",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000003.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000003",
+  "plan_path": "sample_000003/plan.json",
+  "task_path": "sample_000003/vocab_task.json",
+  "prompt_hash": "117c1932f1edb3ce9fdfb3e81dfacb7b09d1402b748b361e9e812444f5375b35",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000003",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000004.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000004",
+  "plan_path": "sample_000004/plan.json",
+  "task_path": "sample_000004/vocab_task.json",
+  "prompt_hash": "6960fecb1d8acad95182b833cb50f6e2533d54c0b80d08771e7bc3b42d40e3d8",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000004",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000005.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000005",
+  "plan_path": "sample_000005/plan.json",
+  "task_path": "sample_000005/vocab_task.json",
+  "prompt_hash": "a48c0aa92c5c1cd47384926ff0c246c81330b81484f9df4241df125f4b568141",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000005",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000006.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000006",
+  "plan_path": "sample_000006/plan.json",
+  "task_path": "sample_000006/vocab_task.json",
+  "prompt_hash": "d570570e4c1353a40b5e3c9c048efe37edf22c7e2a5e5977e66dc49033cbd19d",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000006",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000007.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000007",
+  "plan_path": "sample_000007/plan.json",
+  "task_path": "sample_000007/vocab_task.json",
+  "prompt_hash": "9bdbac530e2f6fdaa0bef9408c2676077dc19d2a8c0de7167594ca7912f19985",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000007",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000008.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000008",
+  "plan_path": "sample_000008/plan.json",
+  "task_path": "sample_000008/vocab_task.json",
+  "prompt_hash": "587eff1134028954e2fb620b54fe1638b5a6fcbd3fe47f6dbdbd07d717459d81",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000008",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000009.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000009",
+  "plan_path": "sample_000009/plan.json",
+  "task_path": "sample_000009/vocab_task.json",
+  "prompt_hash": "a012f98a386adac92d188c68bde72a52c6cbb0dcfd864a08e70d6f08881a15de",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000009",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000010.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000010",
+  "plan_path": "sample_000010/plan.json",
+  "task_path": "sample_000010/vocab_task.json",
+  "prompt_hash": "0ac9cdba25e93fc329b5dd2d566e77d7add5c4dffbf9c53d4f9b8998e0b21917",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000010",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/plan_pool/done/sample_000011.json

@@ -0,0 +1,14 @@
+{
+  "sample_id": "sample_000011",
+  "plan_path": "sample_000011/plan.json",
+  "task_path": "sample_000011/vocab_task.json",
+  "prompt_hash": "9b1f48926b835cfec59e58b4564016b13af9e8f32197b9bfe4a85cddce267178",
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000011",
+  "pool": "plan_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000001.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000001",
+  "plan_path": "sample_000001/plan.json",
+  "task_path": "sample_000001/vocab_task.json",
+  "main_image": "sample_000001/main_image.png",
+  "detections": "sample_000001/detections.json",
+  "references": "sample_000001/references.json",
+  "n_references": 3,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000001",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000002.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000002",
+  "plan_path": "sample_000002/plan.json",
+  "task_path": "sample_000002/vocab_task.json",
+  "main_image": "sample_000002/main_image.png",
+  "detections": "sample_000002/detections.json",
+  "references": "sample_000002/references.json",
+  "n_references": 15,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000002",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000003.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000003",
+  "plan_path": "sample_000003/plan.json",
+  "task_path": "sample_000003/vocab_task.json",
+  "main_image": "sample_000003/main_image.png",
+  "detections": "sample_000003/detections.json",
+  "references": "sample_000003/references.json",
+  "n_references": 3,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000003",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000004.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000004",
+  "plan_path": "sample_000004/plan.json",
+  "task_path": "sample_000004/vocab_task.json",
+  "main_image": "sample_000004/main_image.png",
+  "detections": "sample_000004/detections.json",
+  "references": "sample_000004/references.json",
+  "n_references": 5,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000004",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000005.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000005",
+  "plan_path": "sample_000005/plan.json",
+  "task_path": "sample_000005/vocab_task.json",
+  "main_image": "sample_000005/main_image.png",
+  "detections": "sample_000005/detections.json",
+  "references": "sample_000005/references.json",
+  "n_references": 6,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000005",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000006.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000006",
+  "plan_path": "sample_000006/plan.json",
+  "task_path": "sample_000006/vocab_task.json",
+  "main_image": "sample_000006/main_image.png",
+  "detections": "sample_000006/detections.json",
+  "references": "sample_000006/references.json",
+  "n_references": 8,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000006",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000008.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000008",
+  "plan_path": "sample_000008/plan.json",
+  "task_path": "sample_000008/vocab_task.json",
+  "main_image": "sample_000008/main_image.png",
+  "detections": "sample_000008/detections.json",
+  "references": "sample_000008/references.json",
+  "n_references": 10,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000008",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000009.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000009",
+  "plan_path": "sample_000009/plan.json",
+  "task_path": "sample_000009/vocab_task.json",
+  "main_image": "sample_000009/main_image.png",
+  "detections": "sample_000009/detections.json",
+  "references": "sample_000009/references.json",
+  "n_references": 5,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000009",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000010.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000010",
+  "plan_path": "sample_000010/plan.json",
+  "task_path": "sample_000010/vocab_task.json",
+  "main_image": "sample_000010/main_image.png",
+  "detections": "sample_000010/detections.json",
+  "references": "sample_000010/references.json",
+  "n_references": 9,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000010",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/reference_pool/done/sample_000011.json

@@ -0,0 +1,18 @@
+{
+  "sample_id": "sample_000011",
+  "plan_path": "sample_000011/plan.json",
+  "task_path": "sample_000011/vocab_task.json",
+  "main_image": "sample_000011/main_image.png",
+  "detections": "sample_000011/detections.json",
+  "references": "sample_000011/references.json",
+  "n_references": 10,
+  "reference_errors": {},
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  },
+  "item_id": "sample_000011",
+  "pool": "reference_pool",
+  "retry_count": 0,
+  "errors": []
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/rows/sample_000001.json

@@ -0,0 +1,164 @@
+{
+  "sample_id": "sample_000001",
+  "target_total": 3,
+  "target_people": 1,
+  "target_objects": 2,
+  "canvas_size": [
+    1248,
+    832
+  ],
+  "canvas_aspect_ratio": "3:2",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 3,
+  "n_detected": 3,
+  "n_subjects": 3,
+  "subjects": [
+    {
+      "name": "pedestrian",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_69/273278,12fc4700013112375.jpg:person:3",
+      "source_name": "pedestrian",
+      "source_description": "A person wearing a dark coat and trousers. Source dataset: CrowdHuman. Scene context: A bustling city street lined with trees showcasing vibrant yellow autumn foliage, with many pedestrians walking in both directions.",
+      "sub_caption": "pedestrian: A person wearing a dark coat and trousers.. Scene role: Walking on the sidewalk alongside the street.",
+      "measured_bbox": [
+        0.177,
+        0.0,
+        0.3091,
+        0.4552
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_pedestrian.png",
+      "raw_ref_image": "references/raw_ref_pedestrian_attempt_01.png",
+      "reference_verify": "references/reference_verify_pedestrian.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000001/references/raw_ref_pedestrian_attempt_01.png",
+        "output": "references/ref_pedestrian.png",
+        "mask": "references/sam_mask_pedestrian.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          332.0,
+          40.0,
+          693.0,
+          999.0
+        ],
+        "mask_score": 3.438137,
+        "mask_area_ratio": 0.157722,
+        "elapsed_seconds": 8.8351
+      }
+    },
+    {
+      "name": "parked_dark_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:bcb356f6-520dd65c:object:9",
+      "source_name": "parked dark car",
+      "source_description": "A dark-colored sedan parked on the right side of the street, behind the silver car. Source dataset: BDD100K. Scene context: A large white New York City bus is driving down a city street on an overcast day with other cars parked and driving around it.",
+      "sub_caption": "parked dark car: A dark-colored sedan.. Scene role: Parked on the right side of the street next to the curb.",
+      "measured_bbox": [
+        0.5856,
+        0.0522,
+        0.9973,
+        0.6586
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_parked_dark_car.png",
+      "raw_ref_image": "references/raw_ref_parked_dark_car_attempt_02.png",
+      "reference_verify": "references/reference_verify_parked_dark_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 2,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000001/references/raw_ref_parked_dark_car_attempt_02.png",
+        "output": "references/ref_parked_dark_car.png",
+        "mask": "references/sam_mask_parked_dark_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          17.0,
+          427.0,
+          1006.0,
+          796.0
+        ],
+        "mask_score": 3.312519,
+        "mask_area_ratio": 0.186911,
+        "elapsed_seconds": 8.4991
+      }
+    },
+    {
+      "name": "metal_barrier",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_74/284193,1da20000b642be5b.jpg:object:5",
+      "source_name": "metal barrier",
+      "source_description": "silver metal barricade placed near the entrance Source dataset: CrowdHuman. Scene context: People are gathered outside the entrance of a stone building with arched doorways and large windows.",
+      "sub_caption": "metal barrier: A silver metal barricade.. Scene role: Placed along the edge of the sidewalk near the parked car, separating the walkway from the street.",
+      "measured_bbox": [
+        0.0,
+        0.1355,
+        0.6068,
+        0.558
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_metal_barrier.png",
+      "raw_ref_image": "references/raw_ref_metal_barrier_attempt_01.png",
+      "reference_verify": "references/reference_verify_metal_barrier.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000001/references/raw_ref_metal_barrier_attempt_01.png",
+        "output": "references/ref_metal_barrier.png",
+        "mask": "references/sam_mask_metal_barrier.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          37.0,
+          201.0,
+          1011.0,
+          889.0
+        ],
+        "mask_score": 2.936982,
+        "mask_area_ratio": 0.305722,
+        "elapsed_seconds": 9.8709
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/rows/sample_000002.json

@@ -0,0 +1,716 @@
+{
+  "sample_id": "sample_000002",
+  "target_total": 15,
+  "target_people": 2,
+  "target_objects": 13,
+  "canvas_size": [
+    1280,
+    720
+  ],
+  "canvas_aspect_ratio": "16:9",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 15,
+  "n_detected": 15,
+  "n_subjects": 15,
+  "subjects": [
+    {
+      "name": "pedestrian_right",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "BDD100K:c5d864fa-b0b2380b:person:2",
+      "source_name": "pedestrian",
+      "source_description": "Person walking away from the camera on the right sidewalk, wearing dark clothing. Source dataset: BDD100K. Scene context: View from inside a vehicle driving down a wet city street lined with tall buildings, with pedestrians on the sidewalk and several cars and taxis ahead in the rain.",
+      "sub_caption": "pedestrian: Person walking away from the camera on the right sidewalk, wearing dark clothing.. Scene role: Walking along the right-hand sidewalk under the trees.",
+      "measured_bbox": [
+        0.8872,
+        0.491,
+        0.9451,
+        0.6701
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_pedestrian_right.png",
+      "raw_ref_image": "references/raw_ref_pedestrian_right_attempt_01.png",
+      "reference_verify": "references/reference_verify_pedestrian_right.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_pedestrian_right_attempt_01.png",
+        "output": "references/ref_pedestrian_right.png",
+        "mask": "references/sam_mask_pedestrian_right.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          345.0,
+          55.0,
+          678.0,
+          982.0
+        ],
+        "mask_score": 3.462354,
+        "mask_area_ratio": 0.14014,
+        "elapsed_seconds": 8.2387
+      }
+    },
+    {
+      "name": "pedestrian_left",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "BDD100K:b714a088-861a043b:person:2",
+      "source_name": "pedestrian",
+      "source_description": "another person near the shop entrance on the left, partially obscured Source dataset: BDD100K. Scene context: a city street scene from the perspective of a vehicle, showing multiple cars in motion, buildings lining the road, and pedestrians on the sidewalks during dusk or early evening",
+      "sub_caption": "pedestrian: A person near a shop entrance on the left, partially obscured by shadows.. Scene role: Standing near a building entrance on the left side of the street.",
+      "measured_bbox": [
+        0.1301,
+        0.5154,
+        0.1517,
+        0.611
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_pedestrian_left.png",
+      "raw_ref_image": "references/raw_ref_pedestrian_left_attempt_01.png",
+      "reference_verify": "references/reference_verify_pedestrian_left.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_pedestrian_left_attempt_01.png",
+        "output": "references/ref_pedestrian_left.png",
+        "mask": "references/sam_mask_pedestrian_left.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          342.0,
+          67.0,
+          681.0,
+          996.0
+        ],
+        "mask_score": 3.481605,
+        "mask_area_ratio": 0.150858,
+        "elapsed_seconds": 8.1403
+      }
+    },
+    {
+      "name": "city_buildings",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_9/283554,31eeb000e9237b31.jpg:object:9",
+      "source_name": "building",
+      "source_description": "Various city buildings of different heights forming the skyline in the background. Source dataset: CrowdHuman. Scene context: People are walking and resting on the wooden walkway of a large suspension bridge with a city skyline in the background.",
+      "sub_caption": "building: Various city buildings of different heights forming the urban landscape along the street.. Scene role: Lining the street and forming the architectural background on both sides.",
+      "measured_bbox": [
+        0.3358,
+        0.3425,
+        0.4929,
+        0.5277
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_city_buildings.png",
+      "raw_ref_image": "references/raw_ref_city_buildings_attempt_01.png",
+      "reference_verify": "references/reference_verify_city_buildings.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_city_buildings_attempt_01.png",
+        "output": "references/ref_city_buildings.png",
+        "mask": "references/sam_mask_city_buildings.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          27.0,
+          178.0,
+          996.0,
+          865.0
+        ],
+        "mask_score": 3.420089,
+        "mask_area_ratio": 0.463421,
+        "elapsed_seconds": 8.2735
+      }
+    },
+    {
+      "name": "pink_scooter",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_66/283991,a60c0001c08f63b.jpg:object:3",
+      "source_name": "pink scooter",
+      "source_description": "A prominent pink step-through style motor scooter. Source dataset: CrowdHuman. Scene context: A busy city intersection with many people riding scooters and some cars in the background.",
+      "sub_caption": "pink scooter: A prominent pink step-through style motor scooter.. Scene role: Parked on the right sidewalk near the street signs.",
+      "measured_bbox": [
+        0.677,
+        0.56,
+        0.7935,
+        0.7095
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_pink_scooter.png",
+      "raw_ref_image": "references/raw_ref_pink_scooter_attempt_01.png",
+      "reference_verify": "references/reference_verify_pink_scooter.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_pink_scooter_attempt_01.png",
+        "output": "references/ref_pink_scooter.png",
+        "mask": "references/sam_mask_pink_scooter.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          41.0,
+          120.0,
+          982.0,
+          920.0
+        ],
+        "mask_score": 3.414017,
+        "mask_area_ratio": 0.259921,
+        "elapsed_seconds": 8.1841
+      }
+    },
+    {
+      "name": "street_signs",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b5ab0e46-8eab4733:object:8",
+      "source_name": "street signs",
+      "source_description": "Various street signs attached to a pole on the right side of the street, including a speed limit sign. Source dataset: BDD100K. Scene context: A narrow city street with several parked and moving box trucks and vans, bordered by multi-story buildings and construction barriers.",
+      "sub_caption": "street signs: Various street signs attached to a metal pole on the right side of the street.. Scene role: Mounted on a pole alongside the road on the right.",
+      "measured_bbox": [
+        0.8162,
+        0.2869,
+        0.8575,
+        0.4063
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_signs.png",
+      "raw_ref_image": "references/raw_ref_street_signs_attempt_01.png",
+      "reference_verify": "references/reference_verify_street_signs.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_street_signs_attempt_01.png",
+        "output": "references/ref_street_signs.png",
+        "mask": "references/sam_mask_street_signs.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          226.0,
+          71.0,
+          707.0,
+          967.0
+        ],
+        "mask_score": 3.475593,
+        "mask_area_ratio": 0.25818,
+        "elapsed_seconds": 9.8621
+      }
+    },
+    {
+      "name": "storefront_sign",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_4/273275,46a6f0005d04fc24.jpg:object:7",
+      "source_name": "storefront sign",
+      "source_description": "A dark sign with white text 'DELI' and a smaller red 'ATM' sign beneath it on the left. Source dataset: CrowdHuman. Scene context: A densely crowded city street with numerous pedestrians walking towards the camera, while emergency vehicles with flashing lights are visible in the background traffic.",
+      "sub_caption": "storefront sign: A dark, illuminated sign structure above a shop entrance.. Scene role: Hanging above a shop entrance on the left side of the street, illuminating the adjacent pedestrian.",
+      "measured_bbox": [
+        0.1052,
+        0.4218,
+        0.186,
+        0.4781
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_storefront_sign.png",
+      "raw_ref_image": "references/raw_ref_storefront_sign_attempt_01.png",
+      "reference_verify": "references/reference_verify_storefront_sign.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_storefront_sign_attempt_01.png",
+        "output": "references/ref_storefront_sign.png",
+        "mask": "references/sam_mask_storefront_sign.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          107.0,
+          274.0,
+          900.0,
+          749.0
+        ],
+        "mask_score": 3.354337,
+        "mask_area_ratio": 0.167885,
+        "elapsed_seconds": 8.1782
+      }
+    },
+    {
+      "name": "parked_suv_right",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c54441e6-400c221e:object:4",
+      "source_name": "parked SUV",
+      "source_description": "Dark-colored SUV parked ahead of the sedan on the right side of the road. Source dataset: BDD100K. Scene context: Nighttime driving scene approaching an intersection with green traffic lights and parked cars on the right.",
+      "sub_caption": "parked SUV: Dark-colored SUV parked on the right side of the road.. Scene role: Parked parallel to the curb on the right side of the street.",
+      "measured_bbox": [
+        0.6057,
+        0.5099,
+        0.7451,
+        0.6703
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_parked_suv_right.png",
+      "raw_ref_image": "references/raw_ref_parked_suv_right_attempt_01.png",
+      "reference_verify": "references/reference_verify_parked_suv_right.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_parked_suv_right_attempt_01.png",
+        "output": "references/ref_parked_suv_right.png",
+        "mask": "references/sam_mask_parked_suv_right.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          52.0,
+          216.0,
+          993.0,
+          835.0
+        ],
+        "mask_score": 3.459027,
+        "mask_area_ratio": 0.361156,
+        "elapsed_seconds": 10.1865
+      }
+    },
+    {
+      "name": "dark_car_left",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:bb2e43e4-5e7a7129:object:6",
+      "source_name": "car",
+      "source_description": "A dark car parked along the left curb further ahead. Source dataset: BDD100K. Scene context: A dashcam view from a vehicle driving down a multi-lane road on a partly cloudy day, with a few other cars and pedestrians on the sidewalk.",
+      "sub_caption": "car: A dark car parked along the left curb further ahead.. Scene role: Parked alongside the left curb.",
+      "measured_bbox": [
+        0.2139,
+        0.5323,
+        0.3044,
+        0.6201
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_dark_car_left.png",
+      "raw_ref_image": "references/raw_ref_dark_car_left_attempt_01.png",
+      "reference_verify": "references/reference_verify_dark_car_left.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_dark_car_left_attempt_01.png",
+        "output": "references/ref_dark_car_left.png",
+        "mask": "references/sam_mask_dark_car_left.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          42.0,
+          237.0,
+          982.0,
+          794.0
+        ],
+        "mask_score": 3.479099,
+        "mask_area_ratio": 0.30617,
+        "elapsed_seconds": 8.2274
+      }
+    },
+    {
+      "name": "dark_suv_driving",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c889c950-865ca5b6:object:0",
+      "source_name": "dark SUV",
+      "source_description": "A dark SUV traveling in the left lane, with visible red taillights. Source dataset: BDD100K. Scene context: Nighttime driving on a highway with several cars visible ahead, illuminated mainly by taillights and headlights.",
+      "sub_caption": "dark SUV: A dark SUV traveling in the left lane, with visible glowing red taillights.. Scene role: Driving ahead of the camera vehicle in the opposing or adjacent left lane.",
+      "measured_bbox": [
+        0.3005,
+        0.5101,
+        0.4179,
+        0.6508
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_dark_suv_driving.png",
+      "raw_ref_image": "references/raw_ref_dark_suv_driving_attempt_01.png",
+      "reference_verify": "references/reference_verify_dark_suv_driving.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_dark_suv_driving_attempt_01.png",
+        "output": "references/ref_dark_suv_driving.png",
+        "mask": "references/sam_mask_dark_suv_driving.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          96.0,
+          294.0,
+          928.0,
+          812.0
+        ],
+        "mask_score": 3.455576,
+        "mask_area_ratio": 0.251452,
+        "elapsed_seconds": 9.8494
+      }
+    },
+    {
+      "name": "street_light",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c84f848e-2a5e0737:object:2",
+      "source_name": "street light",
+      "source_description": "Tall pole with a bright light on top, illuminating the road from the right side. Source dataset: BDD100K. Scene context: Nighttime driving view on a multi-lane highway with streetlights illuminating the dark road ahead.",
+      "sub_caption": "street light: Tall pole with a bright light on top, illuminating the road.. Scene role: Casting warm light onto the street from the right-hand sidewalk.",
+      "measured_bbox": [
+        0.8171,
+        0.1755,
+        0.8719,
+        0.2202
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_light.png",
+      "raw_ref_image": "references/raw_ref_street_light_attempt_01.png",
+      "reference_verify": "references/reference_verify_street_light.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_street_light_attempt_01.png",
+        "output": "references/ref_street_light.png",
+        "mask": "references/sam_mask_street_light.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          71.0,
+          277.0,
+          913.0,
+          727.0
+        ],
+        "mask_score": 3.350243,
+        "mask_area_ratio": 0.068855,
+        "elapsed_seconds": 8.2963
+      }
+    },
+    {
+      "name": "vehicle_dashboard",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b5032e1d-dad95b60:object:9",
+      "source_name": "dashboard",
+      "source_description": "The dark, reflective dashboard and lower windshield area of the camera vehicle in the foreground. Source dataset: BDD100K. Scene context: View from inside a vehicle driving down a residential city street lined with trees, parked cars, and multi-story brick buildings.",
+      "sub_caption": "dashboard: The dark, reflective dashboard and lower windshield area of the camera vehicle.. Scene role: Occupying the bottom foreground of the image, establishing the perspective from inside the car.",
+      "measured_bbox": [
+        0.0,
+        0.9261,
+        1.0,
+        1.0
+      ],
+      "detection_confidence": "high",
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_vehicle_dashboard.png",
+      "raw_ref_image": "references/raw_ref_vehicle_dashboard_attempt_01.png",
+      "reference_verify": "references/reference_verify_vehicle_dashboard.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_vehicle_dashboard_attempt_01.png",
+        "output": "references/ref_vehicle_dashboard.png",
+        "mask": "references/sam_mask_vehicle_dashboard.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          24.0,
+          317.0,
+          1001.0,
+          706.0
+        ],
+        "mask_score": 2.942001,
+        "mask_area_ratio": 0.133658,
+        "elapsed_seconds": 8.3645
+      }
+    },
+    {
+      "name": "white_car_ahead",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b3a102ed-6ef54f5e:object:3",
+      "source_name": "white car",
+      "source_description": "A white car visible further down the road in the right lane. Source dataset: BDD100K. Scene context: Nighttime driving scene in a city with cars stopped in traffic and a construction site on the right.",
+      "sub_caption": "white car: A white car visible further down the road.. Scene role: Driving away in the right lane, further in the distance.",
+      "measured_bbox": [
+        0.4811,
+        0.5382,
+        0.5174,
+        0.5915
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_white_car_ahead.png",
+      "raw_ref_image": "references/raw_ref_white_car_ahead_attempt_01.png",
+      "reference_verify": "references/reference_verify_white_car_ahead.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_white_car_ahead_attempt_01.png",
+        "output": "references/ref_white_car_ahead.png",
+        "mask": "references/sam_mask_white_car_ahead.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          112.0,
+          168.0,
+          912.0,
+          855.0
+        ],
+        "mask_score": 3.412999,
+        "mask_area_ratio": 0.338258,
+        "elapsed_seconds": 8.3339
+      }
+    },
+    {
+      "name": "double_yellow_lines",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c417a291-7802692d:object:8",
+      "source_name": "yellow lines",
+      "source_description": "Double yellow painted lines separating opposite directions of traffic. Source dataset: BDD100K. Scene context: A pedestrian crossing a street at a crosswalk with several cars stopped around them, with storefronts visible in the background.",
+      "sub_caption": "yellow lines: Double yellow painted lines separating opposite directions of traffic.. Scene role: Running down the center of the road, receding into the distance.",
+      "measured_bbox": [
+        0.3008,
+        0.5732,
+        0.4776,
+        0.8029
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_double_yellow_lines.png",
+      "raw_ref_image": "references/raw_ref_double_yellow_lines_attempt_01.png",
+      "reference_verify": "references/reference_verify_double_yellow_lines.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_double_yellow_lines_attempt_01.png",
+        "output": "references/ref_double_yellow_lines.png",
+        "mask": "references/sam_mask_double_yellow_lines.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          0.0,
+          1023.0,
+          1023.0
+        ],
+        "mask_score": 2.141169,
+        "mask_area_ratio": 0.667065,
+        "elapsed_seconds": 8.2719
+      }
+    },
+    {
+      "name": "street_trees",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c4891df0-24371ae1:object:3",
+      "source_name": "trees",
+      "source_description": "Numerous trees with dense green and yellowish foliage lining both sides of the road. Source dataset: BDD100K. Scene context: A dark SUV drives ahead on a wet road lined with trees and a fence under a cloudy sky.",
+      "sub_caption": "trees: Numerous trees with dense foliage lining both sides of the road.. Scene role: Planted along the sidewalks, softening the urban environment and framing the street.",
+      "measured_bbox": [
+        0.001,
+        0.002,
+        0.375,
+        0.63
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_trees.png",
+      "raw_ref_image": "references/raw_ref_street_trees_attempt_01.png",
+      "reference_verify": "references/reference_verify_street_trees.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_street_trees_attempt_01.png",
+        "output": "references/ref_street_trees.png",
+        "mask": "references/sam_mask_street_trees.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          65.0,
+          64.0,
+          958.0,
+          969.0
+        ],
+        "mask_score": 3.478968,
+        "mask_area_ratio": 0.365667,
+        "elapsed_seconds": 8.231
+      }
+    },
+    {
+      "name": "twilight_sky",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c13c0d1f-00dfd075:object:9",
+      "source_name": "sky",
+      "source_description": "Clear twilight sky transitioning from bright near the horizon to dark blue at the top. Source dataset: BDD100K. Scene context: View from a moving vehicle looking down a multi-lane city street lined with parked cars and large trees under a clear twilight sky.",
+      "sub_caption": "sky: Clear twilight sky transitioning from bright near the horizon to dark blue at the top.. Scene role: Providing the atmospheric backdrop above the buildings and street.",
+      "measured_bbox": [
+        0.116,
+        0.0,
+        0.714,
+        0.4742
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_twilight_sky.png",
+      "raw_ref_image": "references/raw_ref_twilight_sky_attempt_01.png",
+      "reference_verify": "references/reference_verify_twilight_sky.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000002/references/raw_ref_twilight_sky_attempt_01.png",
+        "output": "references/ref_twilight_sky.png",
+        "mask": "references/sam_mask_twilight_sky.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          208.0,
+          1023.0,
+          814.0
+        ],
+        "mask_score": 2.437955,
+        "mask_area_ratio": 0.529621,
+        "elapsed_seconds": 9.8292
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/rows/sample_000003.json

@@ -0,0 +1,164 @@
+{
+  "sample_id": "sample_000003",
+  "target_total": 3,
+  "target_people": 1,
+  "target_objects": 2,
+  "canvas_size": [
+    1248,
+    832
+  ],
+  "canvas_aspect_ratio": "3:2",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 3,
+  "n_detected": 3,
+  "n_subjects": 3,
+  "subjects": [
+    {
+      "name": "shopper",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_12/282555,5c403000efcca35d.jpg:person:13",
+      "source_name": "shopper",
+      "source_description": "A person standing on the top level, wearing a dark top and dark pants. Source dataset: CrowdHuman. Scene context: A multi-level outdoor shopping mall with various people walking and shopping.",
+      "sub_caption": "shopper: A person standing and waiting, wearing a dark top and dark pants. Scene role: waiting at the crosswalk curb",
+      "measured_bbox": [
+        0.7364,
+        0.2825,
+        0.8267,
+        0.7222
+      ],
+      "detection_confidence": 100,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_shopper.png",
+      "raw_ref_image": "references/raw_ref_shopper_attempt_01.png",
+      "reference_verify": "references/reference_verify_shopper.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000003/references/raw_ref_shopper_attempt_01.png",
+        "output": "references/ref_shopper.png",
+        "mask": "references/sam_mask_shopper.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          348.0,
+          80.0,
+          678.0,
+          995.0
+        ],
+        "mask_score": 3.467753,
+        "mask_area_ratio": 0.132874,
+        "elapsed_seconds": 49.4008
+      }
+    },
+    {
+      "name": "black_sedan",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c0c183ff-1b24f541:object:7",
+      "source_name": "black sedan",
+      "source_description": "A black sedan driving in the right lane ahead of the red hatchback. Source dataset: BDD100K. Scene context: A view from a vehicle driving down a multi-lane city street flanked by tall buildings, with various cars moving in the same direction and parked along the side.",
+      "sub_caption": "black sedan: A dark, modern black sedan. Scene role: driving in the nearest lane on the street",
+      "measured_bbox": [
+        0.0883,
+        0.2514,
+        0.5002,
+        0.449
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_black_sedan.png",
+      "raw_ref_image": "references/raw_ref_black_sedan_attempt_01.png",
+      "reference_verify": "references/reference_verify_black_sedan.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000003/references/raw_ref_black_sedan_attempt_01.png",
+        "output": "references/ref_black_sedan.png",
+        "mask": "references/sam_mask_black_sedan.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          321.0,
+          1023.0,
+          700.0
+        ],
+        "mask_score": 2.52477,
+        "mask_area_ratio": 0.559944,
+        "elapsed_seconds": 8.5091
+      }
+    },
+    {
+      "name": "silver_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:be49ae7a-1ffaa683:object:2",
+      "source_name": "silver car",
+      "source_description": "A silver car driving in the adjacent right lane, slightly ahead of the white car. Source dataset: BDD100K. Scene context: View from inside a vehicle driving in heavy traffic on a multi-lane road under a clear sky.",
+      "sub_caption": "silver car: A sleek silver car. Scene role: driving in the adjacent lane slightly ahead of the black sedan",
+      "measured_bbox": [
+        0.3669,
+        0.2463,
+        0.7048,
+        0.409
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_silver_car.png",
+      "raw_ref_image": "references/raw_ref_silver_car_attempt_01.png",
+      "reference_verify": "references/reference_verify_silver_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000003/references/raw_ref_silver_car_attempt_01.png",
+        "output": "references/ref_silver_car.png",
+        "mask": "references/sam_mask_silver_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          16.0,
+          369.0,
+          1006.0,
+          693.0
+        ],
+        "mask_score": 3.457475,
+        "mask_area_ratio": 0.178123,
+        "elapsed_seconds": 9.7472
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/rows/sample_000004.json

@@ -0,0 +1,256 @@
+{
+  "sample_id": "sample_000004",
+  "target_total": 5,
+  "target_people": 1,
+  "target_objects": 4,
+  "canvas_size": [
+    1248,
+    832
+  ],
+  "canvas_aspect_ratio": "3:2",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 5,
+  "n_detected": 5,
+  "n_subjects": 5,
+  "subjects": [
+    {
+      "name": "pedestrian_walker",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_51/273275,145927000354f7525.jpg:person:10",
+      "source_name": "walker",
+      "source_description": "A man walking, wearing a blue hoodie with 'EMO' written on it, grey pants, and dark shoes. Source dataset: CrowdHuman. Scene context: A group of people, some wearing matching white t-shirts, are walking and jogging along a paved path next to a road, with a grey SUV parked on a grassy hill in the background.",
+      "sub_caption": "walker: A man walking, wearing a plain blue hoodie, grey pants, and dark shoes.. Scene role: Walking near the crosswalk on the side of the street.",
+      "measured_bbox": [
+        0.5948,
+        0.3939,
+        0.6378,
+        0.5698
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_pedestrian_walker.png",
+      "raw_ref_image": "references/raw_ref_pedestrian_walker_attempt_01.png",
+      "reference_verify": "references/reference_verify_pedestrian_walker.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_pedestrian_walker_attempt_01.png",
+        "output": "references/ref_pedestrian_walker.png",
+        "mask": "references/sam_mask_pedestrian_walker.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          334.0,
+          56.0,
+          706.0,
+          996.0
+        ],
+        "mask_score": 3.43302,
+        "mask_area_ratio": 0.160827,
+        "elapsed_seconds": 9.8914
+      }
+    },
+    {
+      "name": "red_traffic_light",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b6df605f-51c158b8:object:6",
+      "source_name": "traffic light",
+      "source_description": "A set of traffic lights suspended over the intersection, showing a red light. Source dataset: BDD100K. Scene context: A view from a vehicle driving down a city street on a sunny day with a U-Haul truck in the opposite lane, parked cars along the right curb, and pedestrians crossing a crosswalk.",
+      "sub_caption": "traffic light: A traffic signal suspended over the intersection, illuminated with a bright red light.. Scene role: Hanging high above the center of the intersection in the driver's line of sight.",
+      "measured_bbox": [
+        0.4668,
+        0.0722,
+        0.5093,
+        0.1896
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_red_traffic_light.png",
+      "raw_ref_image": "references/raw_ref_red_traffic_light_attempt_01.png",
+      "reference_verify": "references/reference_verify_red_traffic_light.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_red_traffic_light_attempt_01.png",
+        "output": "references/ref_red_traffic_light.png",
+        "mask": "references/sam_mask_red_traffic_light.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          286.0,
+          103.0,
+          1023.0,
+          893.0
+        ],
+        "mask_score": 3.25218,
+        "mask_area_ratio": 0.200515,
+        "elapsed_seconds": 8.1927
+      }
+    },
+    {
+      "name": "plain_delivery_truck",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:b3a7b21a-48bcf2b8:object:2",
+      "source_name": "delivery truck",
+      "source_description": "A large white box truck parked behind the fence on the right, with visible green and purple logos. Source dataset: BDD100K. Scene context: A view from a vehicle driving down a wide, paved road flanked by an industrial area with fences and delivery trucks under a cloudy sky.",
+      "sub_caption": "delivery truck: A large, plain white box truck without any visible markings or graphics.. Scene role: Parked alongside the right edge of the street curb.",
+      "measured_bbox": [
+        0.6504,
+        0.2022,
+        0.966,
+        0.6212
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_plain_delivery_truck.png",
+      "raw_ref_image": "references/raw_ref_plain_delivery_truck_attempt_01.png",
+      "reference_verify": "references/reference_verify_plain_delivery_truck.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_plain_delivery_truck_attempt_01.png",
+        "output": "references/ref_plain_delivery_truck.png",
+        "mask": "references/sam_mask_plain_delivery_truck.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          9.0,
+          166.0,
+          1017.0,
+          852.0
+        ],
+        "mask_score": 3.45107,
+        "mask_area_ratio": 0.437578,
+        "elapsed_seconds": 10.0386
+      }
+    },
+    {
+      "name": "dark_parked_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:bcd37eef-1b958ae3:object:5",
+      "source_name": "dark parked car",
+      "source_description": "Another dark-colored car parked on the right side of the street. Source dataset: BDD100K. Scene context: A nighttime view from a vehicle driving down a multi-lane city street with parked cars on the right and oncoming traffic on the left.",
+      "sub_caption": "dark parked car: A dark-colored passenger vehicle.. Scene role: Parked parallel to the curb directly behind the delivery truck.",
+      "measured_bbox": [
+        0.8339,
+        0.4566,
+        0.9965,
+        0.7781
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_dark_parked_car.png",
+      "raw_ref_image": "references/raw_ref_dark_parked_car_attempt_01.png",
+      "reference_verify": "references/reference_verify_dark_parked_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_dark_parked_car_attempt_01.png",
+        "output": "references/ref_dark_parked_car.png",
+        "mask": "references/sam_mask_dark_parked_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          0.0,
+          301.0,
+          1023.0,
+          694.0
+        ],
+        "mask_score": 3.113868,
+        "mask_area_ratio": 0.207836,
+        "elapsed_seconds": 8.5697
+      }
+    },
+    {
+      "name": "street_lines",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:c0c9ec9a-d3638a82:object:6",
+      "source_name": "street lines",
+      "source_description": "Double yellow lines separating traffic directions and white painted lines indicating lanes and crosswalks. Source dataset: BDD100K. Scene context: A view from inside a car driving on a city street, approaching an intersection underneath an elevated railway structure.",
+      "sub_caption": "street lines: Double yellow center lines separating traffic directions, and solid white painted lines forming a distinct crosswalk.. Scene role: Painted on the asphalt, guiding traffic and defining the pedestrian crossing area in the foreground.",
+      "measured_bbox": [
+        0.003,
+        0.432,
+        0.971,
+        0.794
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_lines.png",
+      "raw_ref_image": "references/raw_ref_street_lines_attempt_01.png",
+      "reference_verify": "references/reference_verify_street_lines.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000004/references/raw_ref_street_lines_attempt_01.png",
+        "output": "references/ref_street_lines.png",
+        "mask": "references/sam_mask_street_lines.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          384.0,
+          98.0,
+          639.0,
+          925.0
+        ],
+        "mask_score": 3.44596,
+        "mask_area_ratio": 0.067441,
+        "elapsed_seconds": 8.1646
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}

samples_v8/driving/BDD100K_CrowdHuman_samples/rows/sample_000005.json

@@ -0,0 +1,302 @@
+{
+  "sample_id": "sample_000005",
+  "target_total": 6,
+  "target_people": 3,
+  "target_objects": 3,
+  "canvas_size": [
+    1248,
+    832
+  ],
+  "canvas_aspect_ratio": "3:2",
+  "main_image": "main_image.png",
+  "bbox_overlay": "bbox_overlay.png",
+  "plan": "plan.json",
+  "detections": "detections.json",
+  "vocab_task": "vocab_task.json",
+  "n_planned": 6,
+  "n_detected": 6,
+  "n_subjects": 6,
+  "subjects": [
+    {
+      "name": "person_yellow_top",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_24/282555,49a4e000d44469c6.jpg:person:49",
+      "source_name": "person",
+      "source_description": "Standing, wearing a bright yellow top Source dataset: CrowdHuman. Scene context: A large crowd of tourists is gathered in front of the Louvre museum and its iconic glass pyramid on a sunny day.",
+      "sub_caption": "person: Standing, wearing a bright yellow top. Scene role: Crossing the street on the crosswalk in front of the stopped silver car",
+      "measured_bbox": [
+        0.5309,
+        0.4516,
+        0.5607,
+        0.6301
+      ],
+      "detection_confidence": 0.95,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_person_yellow_top.png",
+      "raw_ref_image": "references/raw_ref_person_yellow_top_attempt_01.png",
+      "reference_verify": "references/reference_verify_person_yellow_top.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_person_yellow_top_attempt_01.png",
+        "output": "references/ref_person_yellow_top.png",
+        "mask": "references/sam_mask_person_yellow_top.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          329.0,
+          42.0,
+          701.0,
+          1012.0
+        ],
+        "mask_score": 3.348943,
+        "mask_area_ratio": 0.150169,
+        "elapsed_seconds": 8.25
+      }
+    },
+    {
+      "name": "man_in_suit",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_28/273278,b62280001bda6f1a.jpg:person:19",
+      "source_name": "crowd member",
+      "source_description": "A person far right in the background wearing a suit. Source dataset: CrowdHuman. Scene context: A crowd of people gathers in front of an old stone building with a prominent arched doorway and two large animal statues.",
+      "sub_caption": "crowd member: A person wearing a professional suit.. Scene role: Walking alongside the other pedestrians across the crosswalk",
+      "measured_bbox": [
+        0.5767,
+        0.4388,
+        0.6397,
+        0.6278
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_man_in_suit.png",
+      "raw_ref_image": "references/raw_ref_man_in_suit_attempt_01.png",
+      "reference_verify": "references/reference_verify_man_in_suit.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_man_in_suit_attempt_01.png",
+        "output": "references/ref_man_in_suit.png",
+        "mask": "references/sam_mask_man_in_suit.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          351.0,
+          27.0,
+          671.0,
+          1004.0
+        ],
+        "mask_score": 3.48496,
+        "mask_area_ratio": 0.144686,
+        "elapsed_seconds": 9.7885
+      }
+    },
+    {
+      "name": "young_girl",
+      "is_person": true,
+      "subject_type": "person",
+      "source_set": "people_set",
+      "source_image_id": "CrowdHuman:data/data_51/282555,57b9f000182048e6.jpg:person:13",
+      "source_name": "pedestrian",
+      "source_description": "Young girl with brown hair, wearing a blue patterned top. Source dataset: CrowdHuman. Scene context: A sunny outdoor scene featuring the red entrance arch to Navy Pier Beer Garden and a tall brick tower, with a diverse crowd of people walking along the waterfront promenade.",
+      "sub_caption": "pedestrian: Young girl with brown hair, wearing a blue patterned top.. Scene role: Walking across the intersection near the person in the yellow top",
+      "measured_bbox": [
+        0.6354,
+        0.4889,
+        0.6677,
+        0.6337
+      ],
+      "detection_confidence": 0.98,
+      "ref_style": "white_bg_full_body_front",
+      "ref_image": "references/ref_young_girl.png",
+      "raw_ref_image": "references/raw_ref_young_girl_attempt_01.png",
+      "reference_verify": "references/reference_verify_young_girl.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_young_girl_attempt_01.png",
+        "output": "references/ref_young_girl.png",
+        "mask": "references/sam_mask_young_girl.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          369.0,
+          52.0,
+          661.0,
+          1003.0
+        ],
+        "mask_score": 3.482282,
+        "mask_area_ratio": 0.133298,
+        "elapsed_seconds": 8.3216
+      }
+    },
+    {
+      "name": "traffic_light",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_36/273275,6a11d000f52c34a9.jpg:object:0",
+      "source_name": "traffic light",
+      "source_description": "A black multi-lens traffic light fixture mounted on a pole above the street. Source dataset: CrowdHuman. Scene context: A male tour guide is speaking to a group of people standing on a city sidewalk next to a road crossing.",
+      "sub_caption": "traffic light: A black multi-lens traffic light fixture mounted on a pole above the street.. Scene role: Suspended over the intersection, showing a red light to halt the vehicles",
+      "measured_bbox": [
+        0.5513,
+        0.0408,
+        0.6462,
+        0.1518
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_traffic_light.png",
+      "raw_ref_image": "references/raw_ref_traffic_light_attempt_01.png",
+      "reference_verify": "references/reference_verify_traffic_light.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_traffic_light_attempt_01.png",
+        "output": "references/ref_traffic_light.png",
+        "mask": "references/sam_mask_traffic_light.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          113.0,
+          201.0,
+          923.0,
+          826.0
+        ],
+        "mask_score": 3.467034,
+        "mask_area_ratio": 0.289252,
+        "elapsed_seconds": 9.874
+      }
+    },
+    {
+      "name": "street_trees",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "CrowdHuman:data/data_37/282555,a4aef000d9c10933.jpg:object:3",
+      "source_name": "trees",
+      "source_description": "Various green trees and shrubs lining the pathway and visible in the background gardens. Source dataset: CrowdHuman. Scene context: A large crowd of tourists walks along the pathway towards the Taj Mahal on a clear day.",
+      "sub_caption": "trees: Various green trees and shrubs lining the pathway and visible in the background gardens.. Scene role: Planted along the sidewalks on both sides of the street, providing urban greenery",
+      "measured_bbox": [
+        0.542,
+        0.2363,
+        0.636,
+        0.493
+      ],
+      "detection_confidence": 0.9,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_street_trees.png",
+      "raw_ref_image": "references/raw_ref_street_trees_attempt_03.png",
+      "reference_verify": "references/reference_verify_street_trees.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 3,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_street_trees_attempt_03.png",
+        "output": "references/ref_street_trees.png",
+        "mask": "references/sam_mask_street_trees.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          22.0,
+          60.0,
+          1003.0,
+          968.0
+        ],
+        "mask_score": 3.301958,
+        "mask_area_ratio": 0.393952,
+        "elapsed_seconds": 8.2223
+      }
+    },
+    {
+      "name": "silver_car",
+      "is_person": false,
+      "subject_type": "object",
+      "source_set": "obj_set",
+      "source_image_id": "BDD100K:be3d3a81-326a032d:object:0",
+      "source_name": "silver car",
+      "source_description": "A silver compact hatchback car facing forward, waiting at an intersection. Its brake lights are on. Source dataset: BDD100K. Scene context: A rainy street scene showing cars waiting at an intersection surrounded by tall buildings and urban infrastructure.",
+      "sub_caption": "silver car: A silver compact hatchback car facing forward, waiting at an intersection. Its brake lights are on.. Scene role: Stopped in the traffic lane in the foreground, waiting for the pedestrians to cross",
+      "measured_bbox": [
+        0.3062,
+        0.4281,
+        0.5436,
+        0.7674
+      ],
+      "detection_confidence": 0.99,
+      "ref_style": "white_bg_encyclopedia_photo",
+      "ref_image": "references/ref_silver_car.png",
+      "raw_ref_image": "references/raw_ref_silver_car_attempt_01.png",
+      "reference_verify": "references/reference_verify_silver_car.json",
+      "reference_verify_passed": true,
+      "reference_attempts": 1,
+      "sam_white_bg": {
+        "input": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/driving/samples/sample_000005/references/raw_ref_silver_car_attempt_01.png",
+        "output": "references/ref_silver_car.png",
+        "mask": "references/sam_mask_silver_car.png",
+        "sam_checkpoint": "/lustre/fs12/portfolios/nvr/projects/nvr_elm_llm/users/hcai/workspace/code/lmlu/codex_gen_data/checkpoints/sam/sam_vit_b_01ec64.pth",
+        "sam_model_type": "vit_b",
+        "sam_device": "auto",
+        "sam_working_size": [
+          640,
+          640
+        ],
+        "sam_max_side": 640,
+        "sam_downscale": 0.625,
+        "prompt_box_xyxy": [
+          80.0,
+          167.0,
+          957.0,
+          937.0
+        ],
+        "mask_score": 3.434142,
+        "mask_area_ratio": 0.414005,
+        "elapsed_seconds": 8.3073
+      }
+    }
+  ],
+  "not_emitted": [],
+  "model_ids": {
+    "chat_model": "gcp/google/gemini-3.1-pro-preview",
+    "image_model": "gcp/google/gemini-3-pro-image-preview"
+  }
+}