Contrastive Decoupled Representation Learning and Regularization for Speech-Preserving Facial Expression Manipulation
Published in International Journal of Computer Vision, Vol. 133, pp. 3822–3838, 2025 · CCF-A, 2025
CDRL introduces contrastive content/emotion representation learning (CCRL/CERL) guided by audio and vision–language priors to reduce content–emotion entanglement and improve lip sync in speech-preserving expression transfer.
Recommended citation: Tianshui Chen, Jianman Lin*, Zhijing Yang, Chunmei Qing, Yukai Shi, Liang Lin. (*Co-first author.) Contrastive Decoupled Representation Learning and Regularization for Speech-Preserving Facial Expression Manipulation. IJCV, 133, 3822–3838, 2025.
Download Paper