BDCNN

the source code of paper, "Block Division Convolutional Network with Implicit Deep Features Augmentation for Micro-Expression Recognition."

Requirements

Python==3.7.6
torch==1.8.1
torchvision==0.9.1
pandas
tqdm
sklearn
matplotlib
opencv_python==4.2.0.34
pickle

Usage

Dataset Preparation

You may put the 3-class and 5-class data into the folder data.
The directory of data shall be represented as follows：

• 006
  • Train
  • img001.img
  • ...
  • label.txt
  • test
  • img001.img
  • ...
  • label.txt
• 007
  ...

Training the BDCNN

Training the 3-class data by using the following command:

cd code  
python train_split.py --model BDCNN --save_path result/BDCNN

Training the 5-class data by using the following command:

python train_5type_split.py  --model BDCNN --save_path result/BDCNN_5type

Get results

Then you can run the result.py to get the result of acc,recall,F1,UF1, and UAR

python result.py  --type 3_class --result_path result/BDCNN

Draw the tsne

python tsne.py  --path result/BDCNN --foder 006 --name BDCNN