MnasNet: Platform-Aware Neural Architecture Search for Mobile
Paper
•
1807.11626
•
Published
SEMnasNet
Use case : Image classification
Model description
SEMnasNet combines the MnasNet architecture with Squeeze-and-Excitation (SE) blocks, adding channel attention mechanisms to the NAS-derived architecture for improved accuracy.
The architecture builds on MnasNet's NAS-derived efficient design and adds Squeeze-and-Excitation blocks for channel attention and feature recalibration. Adaptive feature weighting emphasizes informative channels, with SE blocks boosting accuracy with minimal overhead.
SEMnasNet achieves the highest accuracy in the model zoo (75.38% Top-1) with excellent quantization stability (0.37% drop), making it the best choice for accuracy-critical applications.
(source: https://arxiv.org/abs/1807.11626, https://arxiv.org/abs/1709.01507)
The model is quantized to int8 using ONNX Runtime and exported for efficient deployment.
Network information
| Network Information | Value |
|---|---|
| Framework | Torch |
| MParams | ~4.04 M |
| Quantization | Int8 |
| Provenance | https://github.com/huggingface/pytorch-image-models |
| Paper | https://arxiv.org/abs/1807.11626 |
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 |
|---|---|---|---|---|---|---|---|---|
| semnasnet100_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6 | 2058 | 0 | 4133.38 | 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 |
|---|---|---|---|---|---|---|---|---|
| semnasnet100_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6570-DK | NPU/MCU | 37.63 | 26.57 | 3.0.0 |
| Model | Dataset | Format | Resolution | Board | Execution Engine | Inference time (ms) | Inf / sec | STEdgeAI Core version |
|---|---|---|---|---|---|---|---|---|
| semnasnet100_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6570-DK | NPU/MCU | 37.63 | 26.57 | 3.0.0 |
Accuracy with Imagenet dataset
| Model | Format | Resolution | Top 1 Accuracy |
|---|---|---|---|
| semnasnet100_pt | Float | 224x224x3 | 75.75 % |
| semnasnet100_pt | Int8 | 224x224x3 | 75.38 % |
| Model | Format | Resolution | Top 1 Accuracy |
|---|---|---|---|
| semnasnet100_pt | Float | 224x224x3 | 75.75 % |
| semnasnet100_pt | Int8 | 224x224x3 | 75.38 % |
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 (MnasNet): MnasNet — https://arxiv.org/abs/1807.11626
[3] - Model (SE-Net): Squeeze-and-Excitation Networks — https://arxiv.org/abs/1709.01507