ER-NeRF++: Efficient region-aware Neural Radiance Fields for high-fidelity talking portrait synthesis

Jiahe Li, Jiawei Zhang, Xiao Bai, Jin Zheng, Jun Zhou, Lin Gu

Abstract: Highlights•Facilitate dynamic representation by space pruning with three planar hash encoders.•Build an explicit connection between audio feature and spatial region with attention.•Reuse cross-region points to enhance facial motion via fusing audio-spatial features.•A fast, concise, and effective ER-NeRF for high-fidelity talking portrait synthesis.•An ER-NeRF++ with better quality in facial motion, while maintaining high efficiency.