Related Experiment Video
Updated: Jun 2, 2026
11:34
High-resolution, High-speed, Three-dimensional Video Imaging with Digital Fringe Projection Techniques
Published on: December 3, 2013
15.6K
Towards High-Quality and Disentangled Face Editing in a 3D GAN
Summary
NeRFFaceEditing decouples facial geometry and appearance in neural radiance fields for high-quality, fast editing. This method uses tri-plane statistics and semantic masks for independent control without retraining.
Area of Science:
- Computer Vision
- Computer Graphics
- Artificial Intelligence
Background:
- Neural radiance fields (NeRFs) enable high-quality, fast 3D face synthesis.
- Existing methods struggle with independent editing of geometry and appearance, often requiring retraining.
Purpose of the Study:
- To introduce NeRFFaceEditing for decoupled geometry and appearance editing in pretrained tri-plane NeRFs.
- To maintain high quality and fast inference speeds during editing.
Main Methods:
- Utilizing tri-plane statistics to represent facial volume appearance for disentanglement.
- Employing a 3D-continuous semantic mask as an intermediary for geometry editing.
- Developing separate geometry and appearance decoders with explicit regularization for disentanglement.
Main Results:
- Achieved decoupled control over facial geometry and appearance via semantic masks.
- Demonstrated superior geometry and appearance control compared to existing methods.
- Retained high-quality rendering and fast inference speeds.
Conclusions:
- NeRFFaceEditing successfully enables independent editing of facial geometry and appearance within neural radiance fields.
- The method offers enhanced control and efficiency for 3D face manipulation tasks.
