Multi-task, multi-label and multi-domain learning with residual convolutional networks for emotion recognition

TL;DR

This paper introduces a multi-task learning approach using residual convolutional networks to improve emotion recognition from facial images in challenging, uncontrolled environments, leveraging related tasks like Action Unit detection.

Contribution

It proposes a novel multi-task loss function that effectively models multiple tasks with heterogeneously labeled data, enhancing emotion recognition performance.

Findings

Improved emotion recognition accuracy in wild conditions.

Effective joint learning of emotion and Action Unit detection.

Validation on non-controlled datasets confirms robustness.

Abstract

Automated emotion recognition in the wild from facial images remains a challenging problem. Although recent advances in Deep Learning have supposed a significant breakthrough in this topic, strong changes in pose, orientation and point of view severely harm current approaches. In addition, the acquisition of labeled datasets is costly, and current state-of-the-art deep learning algorithms cannot model all the aforementioned difficulties. In this paper, we propose to apply a multi-task learning loss function to share a common feature representation with other related tasks. Particularly we show that emotion recognition benefits from jointly learning a model with a detector of facial Action Units (collective muscle movements). The proposed loss function addresses the problem of learning multiple tasks with heterogeneously labeled data, improving previous multi-task approaches. We validate the proposal using two datasets acquired in non controlled environments, and an application to predict compound facial emotion expressions.