Domain Generalisation for Apparent Emotional Facial Expression Recognition across Age-Groups

Abstract

Apparent emotional facial expression recognition has attracted a lot of research attention recently. However, the majority of approaches ignore age differences and train a generic model for all ages. In this work, we study the effect of using different age-groups for training apparent emotional facial expression recognition models. To this end, we study Domain Generalisation in the context of apparent emotional facial expression recognition from facial imagery across different age groups. We first compare several domain generalisation algorithms on the basis of out-of-domain-generalisation, and observe that the Class-Conditional Domain-Adversarial Neural Networks (CDANN) algorithm has the best performance. We then study the effect of variety and number of age-groups used during training on generalisation to unseen age-groups and observe that an increase in the number of training age-groups tends to increase the apparent emotional facial expression recognition performance on unseen age-groups. We also show that exclusion of an age-group during training tends to affect more the performance of the neighbouring age groups.

Metrics for Dataset Demographic Bias: A Case Study on Facial Expression Recognition

“…Demographically speaking, some datasets openly focus on facial expressions of specific demographic groups, such as JAFFE [46], [47] (Japanese women) and iSAFE [48] (Indian people), but for the most part the datasets have not been collected taking diversity into account. Some recent works have already found specific biases around gender, race, and age in both commercial FER systems [49], [50] and research models [51], [52], [53], [54], [55], [56]. From these works, Kim et al [49] focus on the age bias of commercial models in an age-labeled dataset.…”

“…Deuschel et al [55] analyzes biases in the prediction of action units with respect to gender and skin color in two popular datasets. Poyiadzy et al [56] work on age bias in a dataset collected from Internet searches, performing additional work to generate apparent age labels for it. Additionally, some works [52], [54], [56] have explored mitigation strategies applicable to this problem.…”

Metrics for Dataset Demographic Bias: A Case Study on Facial Expression Recognition

“…Some recent works have already found specific biases around gender, race, and age in both commercial FER systems [41], [42] and research models [43], [44], [45], [46], [47], [48]. From these works, Kim et al [41] focus on the age bias of commercial models in an age-labeled dataset.…”

“…Deuschel et al [47] analyzes biases in the prediction of action units with respect to gender and skin color in two popular datasets. Poyiadzy et al [48] work on age bias in a dataset collected from Internet searches, performing additional work to generate apparent age labels for it. Additionally, some works [44], [46], [48] have started exploring mitigation strategies applicable to this problem.…”

“…Poyiadzy et al [48] work on age bias in a dataset collected from Internet searches, performing additional work to generate apparent age labels for it. Additionally, some works [44], [46], [48] have started exploring mitigation strategies applicable to this problem. According to the flaws and biases found in previous studies, Hernandez et al [49] propose a set of guidelines to assess and minimize potential risks in the application of FER-based technologies.…”