Micro-Expression Recognition via Fine-Grained Dynamic Perception (TOMM 2025)
Our model is implemented with PyTorch 1.13.0 and Python 3.9.
1.Install requried packages:
$ pip install -r requirements.txt
2.Download Dlib 68_landmarks predictor weight
3.Make training and testing datasets: The datasets like CASME II and SAMM, and should follow such folder structure.
├──CASME2_data_5/
│ ├── disgust
│ │ ├── 01_EP19_05f
│ │ │ ├── img1.jpg
│ │ │ ├── img2.jpg
│ │ │ ├── ......
│ ├── surprise
│ │ ├── ......
├──SAMM_data_5/
│ ├── anger
│ │ ├── 006_1_2
│ │ │ ├── 006_05562.jpg
│ │ │ ├── 006_05563.jpg
│ │ │ ├── ......
│ ├── contempt
│ │ ├── ......
Running data.py makes datasets for each subject:
$ python data.py --dataset SAMM
If you have already made ME datasets, you can simply train FDP like this:
$ python train.py --dataset SAMM