1 Dataset

This code can run on dataset combined of SAMM and CASME II.

All the MERs are divided into 3 classes:

• surprise
  • surprise
• positive
  • happiness
• negative
  • CASME II(disgust, fear, sadness, repression)
  • SAMM(disgust, fear, sadness, repression, anger)

The dataset should be organized as following:

.data_3
├── negative
│   ├── 006_1_2
│   ├── 006_1_3
│   ├── EP01_01
	├── ...
├── positive
│   ├── 007_6_1
│   ├── 007_6_2
│   ├── EP01_01f
	├── ...
├── surprise
│   ├── 006_3_5
│   ├── 006_5_9
│   ├── EP01_13
	├── ...

2 Images

• All frames in the dataset are cropped and resized to 128 * 128.
• If you want to use different size of frames, change the parameter of CNN in MicroExpSTCNN_torch.py.
• You may need to make a dir saved_models_3 to save the models, or you can change the dir name in MicroExpSTCNN_torch.py.