github-luffy/BinaryClassification

BinaryClassify

Implementation of a binary classify By pytorch

DataSets

Prepare a dataset for positive and negative samples. Such as Eye Dataset, determine open eyes as positive samples, closed eyes as negative samples.

1.put positive samples to ./data/positive/

2.put negative samples to ./data/negative/

3.run ./data/train_test_data.py

 $ cd ./data 
 $ python3 train_test_data.py

training & testing

training :

 $ sh train.sh

testing:

 $ python3 test_img.py

pytorch -> onnx -> ncnn

Pytorch -> onnx -> onnx_sim

make sure pip3 install onnx-simplifier

 $ python3 pytorch2onnx.py
 $ python3 -m onnxsim model.onnx model_sim.onnx

onnx_sim -> ncnn

how to build :https://github.com/Tencent/ncnn/wiki/how-to-build

 $ cd ncnn/build/tools/onnx
 $ ./onnx2ncnn model_sim.onnx model_sim.param model_sim.bin

TODO:

• ncnn inference
• train on FocalLoss
• train on multi-class model
• fix bugs