NanoCls

• NanoCls is a simple and lightweight image classification model. And it is very suitable for deployment on embedded or mobile devices. We provide Android demo and MacOS demo based on ncnn inference framework.

Config:mobilenetv2 128x128 width_mult=0.25  batch=128, lr =0.05 
Input Size: torch.Size([1, 3, 128, 128])
MACS： 13.244M  Params: 243.236K
val_accuracy: 0.850 

Config: shufflenetv2 64x64 width_mult=0.5 batch=128, lr =0.05 
Input Size: torch.Size([1, 3, 128, 128]) 
MACS： 13.555M  Params: 345.892K 
val_accuracy: 0.901 

• We provide PyTorch code, and it is very friendly for training with much lower GPU memory cost than other models. We only use grabage classification dataset as tranining dataset.

• We also provide Pytorch code for server deployment

Data

Download password: u94u

cd xxx/xxx/NanoCls
mkdir data
Put your train data into 'data' directory

Train

python train.py 

Test

python predict.py 

PyTorch to ONNX

python pytorch2onnx.py

ONNX to NCNN

https://convertmodel.com/

NanoCls for MacOS

• 1. Modify your own CMakeList.txt
• 2. Build (Apple M1 CPU)

  $ sh make_macos_arm64.sh 
  

NanoCls for Android

Android demo

• 1. Modify your own CMakeList.txt
• 2. Download password: tcmb OpenCV and NCNN libraries for Android

Reference

https://github.com/d-li14/mobilenetv2.pytorch

https://github.com/d-li14/mobilenetv3.pytorch

https://github.com/WZMIAOMIAO/deep-learning-for-image-processing