Xiang Feng, Yongbo He, Yan Yang, Yubo Wang, Yifei Chen, Zhan Wang,Zhenzhong Kuang, Feiwei Qin, Jiajun Ding
Hangzhou Dianzi University.
Overview of the proposed method. SRGS utilizes both external and internal knowledge for joint optimization. The radiance field is optimized in the HR space using the super-splatting method. The external knowledge from the pre-train 2DSR model is used to construct pseudo-GT for high-resolution supervision. The internal knowledge from well-structured existing views facilitates ambiguous Gaussians to learn view-consistent geometry construction.
If you have any questions or suggestions about this repo, please feel free to contact [email protected]