FDMobileNet
Use case : Image classification
Model description
Fast-Downsampling MobileNet (FDMobileNet) is an optimized variant of MobileNet designed for extremely fast inference. It achieves speed improvements through aggressive early spatial reduction while maintaining reasonable accuracy.
FDMobileNet employs a fast downsampling strategy that reduces spatial dimensions early in the network to minimize computation. It retains depthwise separable convolutions inherited from MobileNet for parameter efficiency, and uses a width multiplier (Alpha) to scale the number of channels (a025 = 0.25x, a050 = 0.5x, a075 = 0.75x).
Among the fastest models in the model zoo, FDMobileNet is ideal for ultra-low-latency real-time applications and battery-powered devices with strict power constraints.
(source: https://arxiv.org/abs/1802.03750)
The model is quantized to int8 using ONNX Runtime and exported for efficient deployment.
Network information
| Network Information | Value |
|---|---|
| Framework | Torch |
| MParams | ~0.37–1.77 M |
| Quantization | Int8 |
| Provenance | https://github.com/qinzheng93/FD-MobileNet |
| Paper | https://arxiv.org/abs/1802.03750 |
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 |
|---|---|---|---|---|---|---|---|---|
| fdmobilenet_a025_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6 | 294 | 0 | 377.03 | 3.0.0 |
| fdmobilenet_a050_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6 | 343 | 0 | 973.39 | 3.0.0 |
| fdmobilenet_a075_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6 | 441 | 0 | 1813.66 | 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 |
|---|---|---|---|---|---|---|---|---|
| fdmobilenet_a025_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6570-DK | NPU/MCU | 1.88 | 531.91 | 3.0.0 |
| fdmobilenet_a050_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6570-DK | NPU/MCU | 4.07 | 245.70 | 3.0.0 |
| fdmobilenet_a075_pt_224 | Imagenet | Int8 | 224×224×3 | STM32N6570-DK | NPU/MCU | 6.83 | 146.41 | 3.0.0 |
Accuracy with Imagenet dataset
| Model | Format | Resolution | Top 1 Accuracy |
|---|---|---|---|
| fdmobilenet_a025_pt | Float | 224x224x3 | 45.37 % |
| fdmobilenet_a025_pt | Int8 | 224x224x3 | 29.73 % |
| fdmobilenet_a050_pt | Float | 224x224x3 | 58.04 % |
| fdmobilenet_a050_pt | Int8 | 224x224x3 | 41.58 % |
| fdmobilenet_a075_pt | Float | 224x224x3 | 62.10 % |
| fdmobilenet_a075_pt | Int8 | 224x224x3 | 60.29 % |
| Model | Format | Resolution | Top 1 Accuracy |
|---|---|---|---|
| fdmobilenet_a025_pt | Float | 224x224x3 | 45.37 % |
| fdmobilenet_a025_pt | Int8 | 224x224x3 | 29.73 % |
| fdmobilenet_a050_pt | Float | 224x224x3 | 58.04 % |
| fdmobilenet_a050_pt | Int8 | 224x224x3 | 41.58 % |
| fdmobilenet_a075_pt | Float | 224x224x3 | 62.10 % |
| fdmobilenet_a075_pt | Int8 | 224x224x3 | 60.29 % |
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: FD-MobileNet — https://arxiv.org/abs/1802.03750