This repository contains the code for the paper
Micro-expression recognition with noisy labels, Tuomas Varanka, Wei Peng, Guoying Zhao, IS&T Electronic
Imaging: Human Vision and Electronic Imaging 2021 (157-1-157-8)
link: https://doi.org/10.2352/ISSN.2470-1173.2021.11.HVEI-157
- Setting up the datasets. Go to the datasets.py file and set the paths to the directories of the datasets and provide the excel file path containing the meta data.
- pip install -r requirements.txt
- Change the data paths from datasets.py for
casme2,smic, andsamm. Change both thedf_path, which refers to the excel file containing metadata and thedataset_path, which refers to the cropped dataset root. - Extract optical flow from the datasets and place it in
data/folder. You can also download it from here https://drive.google.com/file/d/1Xhz5qZrJKQn83TJlMKO_hokCA0TkAb7J/view?usp=sharing - Run one of the main*.py files. For simplicity the four different files include the different methods with main.py including the baseline and remaining main*.py consisting of the methods proposed in the paper.
You can change CPU/GPU from the start of the code.
This code uses a pre-computed optical flow using [1], but similar results can be achieved by using Farneback [2] and Dual TV L1 [3] optical flow from OpenCV.
[1] D. Sun, S. Roth, and M. J. Black, “Secrets of optical flow estimation
and their principles,” in International Conference on Computer Vision
and Pattern Recognition (CVPR), 2010, pp. 2432–2439
[2] https://docs.opencv.org/4.x/de/d9e/classcv_1_1FarnebackOpticalFlow.html
[3] https://docs.opencv.org/3.4/dc/d47/classcv_1_1DualTVL1OpticalFlow.html