This repository contains the PyTorch implementation of the core contribution Contextual Causal Intervention Module (CCIM) from the CVPR2023 paper:
Context De-confounded Emotion Recognition
Context-Aware Emotion Recognition (CAER) is a crucial and challenging task that aims to perceive the emotional states of the target person with contextual information. Recent approaches invariably focus on designing sophisticated architectures or mechanisms to extract seemingly meaningful representations from subjects and contexts. However, a long-overlooked issue is that a context bias in existing datasets leads to a significantly unbalanced distribution of emotional states among different context scenarios. Concretely, the harmful bias is a confounder that misleads existing models to learn spurious correlations based on conventional likelihood estimation, significantly limiting the models' performance. To tackle the issue, this paper provides a causality-based perspective to disentangle the models from the impact of such bias, and formulate the causalities among variables in the CAER task via a tailored causal graph. Then, we propose a Contextual Causal Intervention Module (CCIM) based on the backdoor adjustment to de-confound the confounder and exploit the true causal effect for model training. CCIM is plug-in and model-agnostic, which improves diverse state-of-the-art approaches by considerable margins. Extensive experiments on three benchmark datasets demonstrate the effectiveness of our CCIM and the significance of causal insight.
A general pipeline for the context-deconfounded training. The red dotted box shows the core component that achieves the approximation to causal intervention: our CCIM.
To build confounder dictionaries from the different datasets required to implement CCIM, you can access EMOTIC, CAER-S, and GroupWalk to obtain the corresponding training data. Then, the pre-trained backbone on Places365 can be used to extract the context feature set. Ultimately, you can obtain confounder dictionaries of any size by clustering operations.
In CCIM.py, we provide two implementations of dot_product_intervention and additive_intervention. By introducing extracted confounder dictionaries
If you find CCIM useful for your research, please cite the following paper:
@article{yang2023context,
title={Context De-confounded Emotion Recognition},
author={Yang, Dingkang and Chen, Zhaoyu and Wang, Yuzheng and Wang, Shunli and Li, Mingcheng and Liu, Siao and Zhao, Xiao and Huang, Shuai and Dong, Zhiyan and Zhai, Peng and Zhang, Lihua},
journal={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR2023)},
year={2023}
}