In this repository we present our experiments on fer-2013 dataset. Different models like:
- VGG
and different augmentation methods like:
- CutMix
are used in this experiment and more methods are yet to be tried. In case of any errors, please kindly open an issue or create a pull request.
install the requirements using the following command:
pip install -r requirments.txt
Download the official fer2013 dataset and place it in the dataset folder with the following structure datasets/fer2013.csv
The models are compatible with images therefore run the following module to convert fer2013.csv dataset to image files.
python data/csv2img.py --file_path datasets/fer2013.csv --train_path datasets/train --val_path datasets/val --test_path datasets/test
Each training procedure contains a yaml that has the same name of the model to which it belongs, like vgg.yml that belongs to the vgg model.
These config files are required for the training and can be easily modified. Furthermore, to make each training procedure pursuable the yaml file is regenerated in the output directory of
each training.
python train.py --config_path configs/vgg.yml
| Model | Val ACC | Test Acc |
|---|---|---|
| VGG | 70.8833 | 72.0814 |