SA-MobileNetV3

Shuffle Attention for MobileNetV3

Reference

• SA-NET: Shuffle Attention for deep convolutional neural networks: https://github.com/wofmanaf/SA-Net

Experiments

on ImageNet

Attempt	Parameters	Madds	Top1-acc	Sample visualization
MobileNetV3-Large	5.4 M	448.69 M	75.2%
SA-MobileNetV3-Large	3.9 M	445.68 M	76.8%

on CIFAR-10

Attempt	Parameters	Madds	Top1-acc	Sample visualization
MobileNetV3-Large	4.2 M	446.16 M
SA-MobileNetV3-Large	2.7 M	443.14 M

on MNist

Attempt	Parameters	Madds	Top1-acc	Sample visualization	Sample visualization
MobileNetV3-Large	4.2 M	446.16 M	0.997%
SA-MobileNetV3-Large	2.7 M	443.14 M	0.998%

Train

Run the following command for train model on your own dataset:

python train.py --dataset mnist 

Test

Run the following command for evaluation trained model on test dataset:

python test.py --dataset mnist

Inference

Run the following command for classification images:

python inference.py --input /path/to/image.jpg 

Citation

Please cite our paper if you find this repo useful in your research.

@article{
}