ReCoT: Regularized Co-Training for Facial Action Unit Recognition with Noisy Labels
Yifan Li (Michigan State University), Hu Han (Institute of Computing Technology, Chinese Academy of Sciences),* Shiguang Shan (Institute of Computing Technology, Chinese Academy of Sciences), zhilong ji (Tomorrow Advancing Life), Jinfeng Bai (Tomorrow Advance Life), Xilin Chen (Institute of Computing Technology, Chinese Academy of Sciences)
The 34th British Machine Vision Conference
Abstract
Facial action unit (AU) recognition is essential for recognizing fine-grained changes in facial expression, while the demand for a large amount of accurately labeled AU data for training purposes has resulted in high labor costs. Nevertheless, massive face images are widely available and inaccurate labels can be easily obtained, especially as large vision-language pre-training models progress. This paper introduces the Regularized Co-Training (ReCoT) method, which leverages the useful information from both accurately labeled (clean) and inaccurately labeled (noisy) face images to achieve robust AU recognition. ReCoT uses a two-head network in each view, with one for clean data modeling (clean net) and the other for noisy data modeling (noisy net) by learning label noise w.r.t. the clean predictions. Additionally, a selective balanced loss is proposed for the noisy net to learn from noisy labels and alleviate the imbalanced issue in the clean net. Extensive experiments on several AU databases, including EmotioNet, BP4D and DISFA, show that ReCoT effectively leverages noisy AU data to improve the model performance.Citation
@inproceedings{Li_2023_BMVC,
author = {Yifan Li and Hu Han and Shiguang Shan and zhilong ji and Jinfeng Bai and Xilin Chen},
title = {ReCoT: Regularized Co-Training for Facial Action Unit Recognition with Noisy Labels},
booktitle = {34th British Machine Vision Conference 2023, {BMVC} 2023, Aberdeen, UK, November 20-24, 2023},
publisher = {BMVA},
year = {2023},
url = {https://papers.bmvc2023.org/0102.pdf}
}
Copyright © 2023
The British Machine Vision Conference is organised by