FaceCanvas: Action Unit Palette for Facial Transformations

Abstract

This talk introduces FaceCanvas, a project on fine-grained facial expression generation and editing using facial Action Units (AUs). Instead of relying on coarse emotion labels such as “happy” or “sad”, FaceCanvas aims to provide precise and interpretable control over facial movements through AU combinations and intensity levels. I will briefly present the project goals, including AU-conditioned image/video generation, micro-expression generation, multi-modal control, and responsible use, as well as recent progress and preliminary results from our ongoing work.

Bio

Mengting Wei is now a PhD student in University of Oulu where she works in the Center for Machine Vision and Signal Analysis (CMVS) under the supervision of Prof. Guoying Zhao. She is also a PhD student in European Laboratory for Learning and Intelligent Systems (ELLIS), co-supervised by Prof. Nuria Oliver. Previously, she graduated from Southeast University with a master degree, supervised by Prof. Wenming Zheng. She received her bechelor degree from Hebei University of Technology. Her current research interests lie at face generation and editing.