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

Dynamic facial expression recognition (DFER) is essential to the developmentof intelligent and empathetic machines. Prior efforts in this field mainly fallinto supervised learning paradigm, which is severely restricted by the limitedlabeled data in existing datasets. Inspired by recent unprecedented success ofmasked autoencoders (e.g., VideoMAE), this paper proposes MAE-DFER, a novelself-supervised method which leverages large-scale self-supervised pre-trainingon abundant unlabeled data to largely advance the development of DFER. Sincethe vanilla Vision Transformer (ViT) employed in VideoMAE requires substantialcomputation during fine-tuning, MAE-DFER develops an efficient local-globalinteraction Transformer (LGI-Former) as the encoder. Moreover, in addition tothe standalone appearance content reconstruction in VideoMAE, MAE-DFER alsointroduces explicit temporal facial motion modeling to encourage LGI-Former toexcavate both static appearance and dynamic motion information. Extensiveexperiments on six datasets show that MAE-DFER consistently outperformsstate-of-the-art supervised methods by significant margins (e.g., +6.30\% UARon DFEW and +8.34\% UAR on MAFW), verifying that it can learn powerful dynamicfacial representations via large-scale self-supervised pre-training. Besides,it has comparable or even better performance than VideoMAE, while largelyreducing the computational cost (about 38\% FLOPs). We believe MAE-DFER haspaved a new way for the advancement of DFER and can inspire more relevantresearch in this field and even other related tasks. Codes and models arepublicly available at https://github.com/sunlicai/MAE-DFER.

MAE-DFER: Efficient Masked Autoencoder for Self-supervised Dynamic Facial Expression Recognition