OSNet

Use case : Re-Identification

Model description

OSNet is a lightweight convolutional neural network architecture designed specifically for person re-identification tasks. It introduces omni-scale feature learning, enabling the network to capture multi-scale information efficiently within a single residual block.

Key features of OSNet:

• Omni-scale feature learning for robust representation.
• Lightweight design with fewer parameters compared to traditional re-identification models.
• Suitable for deployment on resource-constrained devices.

For more details, see the OSNet paper: https://arxiv.org/abs/1905.00953

The model is quantized using ONNX quantization tools.

Network information

Network Information	Value
Framework	TensorFlow Lite
MParams alpha=0.25	0.197 M
Quantization	int8
Provenance	https://kaiyangzhou.github.io/deep-person-reid/index.html
Paper	https://arxiv.org/abs/1905.0095

The models are quantized using TF Lite post-training quantization tools.

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	[x]	[]
STM32U5	[x]	[]
STM32H7	[x]	[x]
STM32MP1	[x]	[x]
STM32MP2	[x]	[x]
STM32N6	[x]	[x]

Performances

Metricss

• Measures are done with default STM32Cube.AI configuration with enabled input / output allocated option.
• tfs stands for "training from scratch", meaning that the model weights were randomly initialized before training.
• tl stands for "transfer learning", meaning that the model backbone weights were initialized from a pre-trained model, then only the last layer was unfrozen during the training.
• fft stands for "full fine-tuning", meaning that the full model weights were initialized from a transfer learning pre-trained model, and all the layers were unfrozen during the training.

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

Model	Dataset	Format	Resolution	Series	Internal RAM	External RAM	Weights Flash	STEdgeAI Core version
OSNet 0.25 tfs	DeepSportradar	Int8	256x128x3	STM32N6	480	0	404.94	3.0.0
OSNet 1.0 tfs	DeepSportradar	Int8	256x128x3	STM32N6	1440	0	2375.33	3.0.0

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

Model	Dataset	Format	Resolution	Board	Execution Engine	Inference time (ms)	Inf / sec	STEdgeAI Core version
OSNet 0.25 tfs	DeepSportradar	Int8	256x128x3	STM32N6570-DK	NPU/MCU	3.53	283.3	3.0.0
OSNet 1.0 tfs	DeepSportradar	Int8	256x128x3	STM32N6570-DK	NPU/MCU	13.44	74.4	3.0.0

Reference MCU memory footprint based on DeepSportradar dataset (see Accuracy for details on dataset)

Model	Dataset	Format	Resolution	Series	Internal RAM	External RAM	Weights Flash	STEdgeAI Core version
OSNet 0.25 tfs	DeepSportradar	Int8	256x128x3	STM32H7	331.45	0	139.52	3.0.0
OSNet 1.0 tfs	DeepSportradar	Int8	256x128x3	STM32H7	396.01	1024.0	1892.75	3.0.0

Reference MCU inference time on DeepSportradar dataset (see Accuracy for details on dataset)

Model	Dataset	Format	Resolution	Board	Execution Engine	Inference time (ms)	Inf / sec	STEdgeAI Core version
OSNet 0.25 tfs	DeepSportradar	Int8	256x128x3	STM32H747I-DISCO	1 CPU	495.13	2.02	3.0.0
OSNet 1.0 tfs	DeepSportradar	Int8	256x128x3	STM32H747I-DISCO	1 CPU	3894.82	0.26	3.0.0

Performance with DeepSportradar ReID dataset

Dataset details: link , License Apache-2.0 , Number of identities: 486 (train: 436, test: 50), Number of images: 9529 (train: 8569, test_query: 50, test_gallery: 910)

Model	Format	Resolution	mAP	rank-1 accuracy	rank-5 accuracy	rank-10 accuracy
OSNet 0.25 tfs	Int8	256x128	70.27 %	92.0 %	96.0 %	96.0 %
OSNet 1.0 tfs	Int8	256x128	73.84 %	90.0 %	98.0 %	98.0 %

Retraining and Integration in a simple example:

Please refer to the stm32ai-modelzoo-services GitHub here

References

[1] The DeepSportradar Player Re-Identification Challenge (2023) [Online]. Available: https://github.com/DeepSportradar/player-reidentification-challenge.