STMicroelectronics/mobilenetv2_pt

MobileNet V2

Use case : Image classification

Model description

MobileNetV2 improves upon V1 with inverted residual blocks and linear bottlenecks. It features skip connections between thin bottleneck layers, improving gradient flow and enabling deeper, more accurate networks.

The architecture uses inverted residuals that expand channels before depthwise convolution and then compress, with linear bottlenecks that remove non-linearity in narrow layers to preserve information. Skip connections between bottleneck layers improve gradient flow, typically with a 6x expansion ratio.

MobileNetV2 offers the best overall accuracy-efficiency trade-off with excellent quantization stability (typically <1% accuracy drop), making it ideal for production mobile applications, object detection backbones, and semantic segmentation networks.

(source: https://arxiv.org/abs/1801.04381)

The model is quantized to int8 using ONNX Runtime and exported for efficient deployment.

Network information

Network Information	Value
Framework	Torch
MParams	~1.49–3.72 M
Quantization	Int8
Provenance	https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet
Paper	https://arxiv.org/abs/1801.04381

Network inputs / outputs

For an image resolution of NxM and P classes

Input Shape	Description
(1, N, M, 3)	Single NxM RGB image with UINT8 values between 0 and 255

Output Shape	Description
(1, P)	Per-class confidence for P classes in FLOAT32

Recommended platforms

Platform	Supported	Recommended
STM32L0	[]	[]
STM32L4	[]	[]
STM32U5	[]	[]
STM32H7	[]	[]
STM32MP1	[]	[]
STM32MP2	[]	[]
STM32N6	[x]	[x]

Performances

Metrics

• Measures are done with default STEdgeAI Core configuration with enabled input / output allocated option.
• All the models are trained from scratch on Imagenet dataset

Reference NPU memory footprint on Imagenet dataset (see Accuracy for details on dataset)

Model	Dataset	Format	Resolution	Series	Internal RAM (KiB)	External RAM (KiB)	Weights Flash (KiB)	STEdgeAI Core version
mobilenetv2_a025_pt_224	Imagenet	Int8	224×224×3	STM32N6	392	0	1522.31	3.0.0
mobilenetv2b_a025_pt_224	Imagenet	Int8	224×224×3	STM32N6	392	0	1522.25	3.0.0
mobilenetv2_w035_pt_224	Imagenet	Int8	224×224×3	STM32N6	931	0	1685.00	3.0.0
mobilenetv2_a050_pt_224	Imagenet	Int8	224×224×3	STM32N6	1274	0	1972.03	3.0.0
mobilenetv2b_a050_pt_224	Imagenet	Int8	224×224×3	STM32N6	1065.75	0	1965.86	3.0.0
mobilenetv2_a075_pt_224	Imagenet	Int8	224×224×3	STM32N6	1653.75	0	2737.58	3.0.0
mobilenetv2b_a075_pt_224	Imagenet	Int8	224×224×3	STM32N6	1653.75	0	2737.02	3.0.0
mobilenetv2_a100_pt_224	Imagenet	Int8	224×224×3	STM32N6	2058	0	3813.97	3.0.0
mobilenetv2b_a100_pt_224	Imagenet	Int8	224×224×3	STM32N6	2058	0	3812.52	3.0.0

Reference NPU inference time on Imagenet dataset (see Accuracy for details on dataset)

Model	Dataset	Format	Resolution	Board	Execution Engine	Inference time (ms)	Inf / sec	STEdgeAI Core version
mobilenetv2_a025_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	6.50	153.85	3.0.0
mobilenetv2_a050_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	10.08	99.21	3.0.0
mobilenetv2_a075_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	15.17	65.88	3.0.0
mobilenetv2_a100_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	20.35	49.14	3.0.0
mobilenetv2_w035_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	8.58	116.55	3.0.0
mobilenetv2b_a025_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	6.29	158.98	3.0.0
mobilenetv2b_a050_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	9.79	102.14	3.0.0
mobilenetv2b_a075_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	14.56	68.68	3.0.0
mobilenetv2b_a100_pt_224	Imagenet	Int8	224×224×3	STM32N6570-DK	NPU/MCU	20.39	49.04	3.0.0

Accuracy with Imagenet dataset

model	Format	Resolution	Top 1 Accuracy
mobilenetv2_a025_pt	Float	224x224x3	52.29 %
mobilenetv2_a025_pt	Int8	224x224x3	51.51 %
mobilenetv2_a050_pt	Float	224x224x3	66.20 %
mobilenetv2_a050_pt	Int8	224x224x3	65.31 %
mobilenetv2_a075_pt	Float	224x224x3	70.78 %
mobilenetv2_a075_pt	Int8	224x224x3	70.33 %
mobilenetv2_a100_pt	Float	224x224x3	73.17 %
mobilenetv2_a100_pt	Int8	224x224x3	72.76 %
mobilenetv2_w035_pt	Float	224x224x3	61.02 %
mobilenetv2_w035_pt	Int8	224x224x3	60.09 %
mobilenetv2b_a025_pt	Float	224x224x3	53.53 %
mobilenetv2b_a025_pt	Int8	224x224x3	52.55 %
mobilenetv2b_a050_pt	Float	224x224x3	66.30 %
mobilenetv2b_a050_pt	Int8	224x224x3	65.67 %
mobilenetv2b_a075_pt	Float	224x224x3	70.41 %
mobilenetv2b_a075_pt	Int8	224x224x3	70.20 %
mobilenetv2b_a100_pt	Float	224x224x3	73.33 %
mobilenetv2b_a100_pt	Int8	224x224x3	72.89 %

model	Format	Resolution	Top 1 Accuracy
mobilenetv2_a025_pt	Float	224x224x3	52.29 %
mobilenetv2_a025_pt	Int8	224x224x3	51.51 %
mobilenetv2_a050_pt	Float	224x224x3	66.20 %
mobilenetv2_a050_pt	Int8	224x224x3	65.31 %
mobilenetv2_a075_pt	Float	224x224x3	70.78 %
mobilenetv2_a075_pt	Int8	224x224x3	70.33 %
mobilenetv2_a100_pt	Float	224x224x3	73.17 %
mobilenetv2_a100_pt	Int8	224x224x3	72.76 %
mobilenetv2_w035_pt	Float	224x224x3	61.02 %
mobilenetv2_w035_pt	Int8	224x224x3	60.09 %
mobilenetv2b_a025_pt	Float	224x224x3	53.53 %
mobilenetv2b_a025_pt	Int8	224x224x3	52.55 %
mobilenetv2b_a050_pt	Float	224x224x3	66.30 %
mobilenetv2b_a050_pt	Int8	224x224x3	65.67 %
mobilenetv2b_a075_pt	Float	224x224x3	70.41 %
mobilenetv2b_a075_pt	Int8	224x224x3	70.20 %
mobilenetv2b_a100_pt	Float	224x224x3	73.33 %
mobilenetv2b_a100_pt	Int8	224x224x3	72.89 %

Retraining and Integration in a simple example:

Please refer to the stm32ai-modelzoo-services GitHub here

References

[1] - Dataset: Imagenet (ILSVRC 2012) — https://www.image-net.org/

[2] - Model: MobileNetV2 — https://github.com/tensorflow/models/tree/master/research/slim/nets/mobilenet