An End-to-end Learning-based Approach to 3D Novel View Style Transfer
Kai-Cheng Chang, Yi‐Ping Hung, Chu‐Song Chen · 2022
3D novel view style transfer is a rising research topic. Recently developed methods aim to build globally optimized scene representations and stylize them directly on the scene. However, these methods are time-consuming because they need globally-consistent optimization or rendering fields reconstruction. In this paper, we introduce an end-to-end learning framework to handle the problem of stylized novel view synthesis, which can speed up the 3D style transfer by applying learning-based structure-of-motion (SfM) approaches. Experimental results show that our method can achieve comparable visual effects to the original style transfer module with higher efficiency.