Learned Initializations for Optimizing Coordinate-Based Neural Representations

2020-12-03 2020-12-03 约 162 字预计阅读 1 分钟

Learned initializations for optimizing coordinate-based neural representations

CVPR2021 Proceedings of the IEEE/CVF conference on computer vision and pattern recognition

Matthew Tancik, Ben Mildenhall, Terrance Wang, Divi Schmidt, Pratul P Srinivasan, Jonathan T Barron, Ren Ng

UCB Google

meta-learning

编者按

对于coordinate-based neural representations在auto-decoder时，用meta-learned 的initialization
与MetaSDF的差别：进一步拓展到更多种类的neural coordinate-based signals，并且把the power of using initial weight settings开发为一种先验信息