High-fidelity and realtime 3D Gaussian Head Avatars with Expressive and Compact blenshape representations

Yating Wang, Xuan Wang, Ran Yi, Yanbo Fan, Jichen Hu, Jingcheng Zhu, Lizhuang Ma

Keywords: head avatar, gaussian splatting

Abstract:

Recent studies have combined 3D Gaussian and 3D Morphable Models (3DMM) to achieve real-time, high-quality rendering of controllable head avatars. Several techniques have attempted to express dynamic textures in facial animation when modeling 3D avatars. However, accurately capturing and displaying expressive appearance dynamics while maintaining temporal and spatial efficiency remains a technical challenge. To this end, we propose a novel method for 3D facial avatar modeling that utilizes an expressive and compact model representation, capturing dynamic facial information accurately while ensuring efficiency. We encode texture-related attributes of the 3D Gaussians in the tensorial feature representation. Specifically, we store color information of the neutral expression in static tri-planes; and represent dynamic texture details for different expressions using lightweight 1D feature lines, which are then decoded into opacity changes relative to the neutral face. Experiments show that this design introduces nonlinear expressiveness to the model, enhancing its performance, while the compact representation maintains real-time rendering capabilities and significantly reduces storage costs. This approach thus broadens the applicability to more scenarios.

Primary Area: applications to computer vision, audio, language, and other modalities

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 877