To run FERPlus, please follow several steps:
Step 1: cd into FERPlus directory
Step 2: generate 2 json files
python3 get_json_DIS.py # generate info_DIS.json
python3 get_json_MV.py #generate info_MV.json
These two json files include all data about FERPlus,so they are very large (~8G in total) and it takes approximately 18 minutes to generate them.We mainly use info_DIS.json. The function of two json files are listed below:
| json file | usage |
|---|---|
| info_DIS.json | images with probability distribution tags |
| info_MV.json | images with one-hot tags |
Step3:
python3 main_lr_0.0001.py (initial learning rate=0.0001), or
python3 main_lr_0.001.py (initial learning rate=0.001)
Organized datasets and two json files could be downloaded here (info.zip):
url: https://pan.baidu.com/s/1adg0JLiMkDb7YMLZe71yJQ
password: dfig
or connect us: miaosi2018@sari.ac.cn, 2904661326@qq.com, miaosi@hust.edu.cn (miaosi2018@sari.ac.cn is recommended
Step1: Download "Cropped" from Xiaolan Fu's website. If you need our original preprocessed optical flow, please send a cc that Xiaolan Fu has agreed your application for CASME2. The CASME2-coding-20140508.csv file and the "Cropped" directory are intentionally left blank due to the license.
Step2: run python3 calcflow.py to generate optical flow and info_CASME2.json (It is about some information on the optical flow images).
Step3: run python3 call.py. It would run LOSO 26 times.