The official implementation of the following paper at NeurIPS2021:
Title: Relative Uncertainty Learning for Facial Expression Recognition
Authors: Yuhang Zhang, Chengrui Wang, Weihong Deng
Institute: BUPT
In facial expression recognition (FER), the uncertainties introduced by inherent noises like ambiguous facial expressions and inconsistent labels raise concerns about the credibility of recognition results. To quantify these uncertainties and achieve good performance under noisy data, we regard uncertainty as a relative concept and propose an innovative uncertainty learning method called Relative Uncertainty Learning (RUL). Rather than assuming Gaussian uncertainty distributions for all datasets, RUL builds an extra branch to learn uncertainty from the relative difficulty of samples by feature mixup. Specifically, we use uncertainties as weights to mix facial features and design an add-up loss to encourage uncertainty learning. It is easy to implement and adds little or no extra computation overhead. Extensive experiments show that RUL outperforms state-of-the-art FER uncertainty learning methods in both real-world and synthetic noisy FER datasets. Besides, RUL also works well on other datasets such as CIFAR and Tiny ImageNet.
The feature distribution figure shows that RUL encourages intra-class compactness and inter-class seperability of the learned features. (0:Surprise, 1:Fear, 2:Disgust, 3:Happy, 4:Sad, 5:Angry, 6:Neutral)
Torch
We train RUL with Torch 1.8.0 and torchvision 0.9.0.
Dataset
Download RAF-DB, put it into the dataset folder, and make sure that it has the same structure as bellow:
- dataset/raf-basic/
EmoLabel/
list_patition_label.txt
Image/aligned/
train_00001_aligned.jpg
test_0001_aligned.jpg
...
Pretrained backbone model
Download the pretrained ResNet18 from this github repository, and then put it into the pretrained_model directory. We thank the authors for providing their pretrained ResNet model.
The pre-trained model of the mentioned github repository is expired. You can acquire the pre-trained model through this link.
Train the RUL model
cd src
python main.py --raf_path '../dataset/raf-basic' --label_path '../dataset/raf-basic/EmoLabel/list_patition_label.txt' --pretrained_backbone_path '../pretrained_model/resnet18_msceleb.pth'
Citation
If you find our code useful, please consider citing our paper
@article{zhang2021relative,
title={Relative Uncertainty Learning for Facial Expression Recognition},
author={Zhang, Yuhang and Wang, Chengrui and Deng, Weihong},
journal={Advances in Neural Information Processing Systems},
volume={34},
pages={17616--17627},
year={2021}
}Acknowledgments
Our work is based on the following works, thanks for their code and pretrained model:
https://github.com/kaiwang960112/Self-Cure-Network