English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Studies have proven that domain bias and label bias exist in different FacialExpression Recognition (FER) datasets, making it hard to improve theperformance of a specific dataset by adding other datasets. For the FER biasissue, recent researches mainly focus on the cross-domain issue with advanceddomain adaption algorithms. This paper addresses another problem: how to boostFER performance by leveraging cross-domain datasets. Unlike the coarse andbiased expression label, the facial Action Unit (AU) is fine-grained andobjective suggested by psychological studies. Motivated by this, we resort tothe AU information of different FER datasets for performance boosting and makecontributions as follows. First, we experimentally show that the naive jointtraining of multiple FER datasets is harmful to the FER performance ofindividual datasets. We further introduce expression-specific mean images andAU cosine distances to measure FER dataset bias. This novel measurement showsconsistent conclusions with experimental degradation of joint training. Second,we propose a simple yet conceptually-new framework, AU-aware Vision Transformer(AU-ViT). It improves the performance of individual datasets by jointlytraining auxiliary datasets with AU or pseudo-AU labels. We also find that theAU-ViT is robust to real-world occlusions. Moreover, for the first time, weprove that a carefully-initialized ViT achieves comparable performance toadvanced deep convolutional networks. Our AU-ViT achieves state-of-the-artperformance on three popular datasets, namely 91.10% on RAF-DB, 65.59% onAffectNet, and 90.15% on FERPlus. The code and models will be released soon.

AU-Aware Vision Transformers for Biased Facial Expression Recognition