JING-MING GUO1, (Senior Member, IEEE), CHIH-HSIEN HSIA2, (Member, IEEE), PING-HSUEH HO1, (Bachelor), YANG-CHEN CHANG1, (Bachelor)
1Department of Electrical Engineering, National Taiwan University of Science and Technology Taipei City 106, Taipei County, Taiwan 2Department of Computer Science and Information Engineering, National Ilan University, Yilan City 260, Yilan County, Taiwan
1jmguo@mail.ntust.edu.tw, bernie6401@gmail.com, Max.chang965132@gmail.com 2chhsia625@gmail.com
The datasets are placed in the datasets folder, we prepare them as the link you can download by correct structure, please see the data readme.
In the folder, we just use GEMEP dataset that has whole body including face and body especially as our research data and compare with other paper.
The modes' weights are placed in the models folder, we prepare them as the link you can download by correct structure, please see model readme.
We use Fusing Body Posture model as our body model and also use ARM model as our face model. We also use weights that ARM team(named epoch59_acc0.9205.pth in download link) and Fusing Body team provided as our pretrained weight respectively.
Our model structure is as below:
Setup environment:
pip install -U scikit-learn
conda install -c conda-forge matplotlib
conda install -c conda-forge argparse
conda install -c conda-forge tqdm
conda install -c conda-forge wandb
conda install -c anaconda pillow
Demo GEMEP with 7 classes
python test_leave_one_out.py -p --data_type GEMEP --num_classes 7
Demo GEMEP with 12 classes
python test_leave_one_out.py -p --data_type GEMEP --num_classes 12
You can download here for more experience detail such as platform, software version, or hyper parameters, etc.
- 2022/08/14 update dataset_leave_one_out.py and test_leave_one_out.py files about adding emo_tranform() function in former file to parse num_classes argument and let the user used it more convenient. Now you can just use the comment above and no need to adjust any code in any file in testing mode.