This work is the final project of the Computer Vision Course of USTC. However, I achieve the highest single-network classification accuracy on FER2013 based on ResNet18. To my best knowledge, this work achieves state-of-the-art single-network accuracy of 73.70 % on FER2013 without using extra training data, which exceeds the previous work [1] of 73.28%. (Chineses Post)
Method | Private Test Data |
---|---|
[1] | 73.28% |
This work | 73.70% |
Official model checkpoint and training log can be found following:
Fer2013 Leaderboard: Here
- GPU:2080Ti
- py:37
其他
- CUDA:10.2
- cuDnn:7605
First, you should download the
official fer2013
dataset, and place it in the outmost folder with the following folder structure datasets/fer2013/fer2013.csv
To train your own model, run the following:
python train.py --name='your_version'
To evaluate the model, run the follwing
python evaluate.py --checkpoint='xxx/best_checkpoint.tar'
[1] Khaireddin, Yousif, and Zhuofa Chen. "Facial Emotion Recognition: State of the Art Performance on FER2013." arXiv preprint arXiv:2105.03588 (2021).