CVPR2020
Seeing the World in a Bag of Chips
Jeong Joon Park, Aleksander Holynski, Steven M. Seitz
摘要
Input images (b) Estimated environment (top), ground truth (bottom) (c) Zoom-in Figure 1: From a hand-held RGBD sequence of an object (a), we reconstruct an image of the surrounding environment (b, top) that closely resembles the real environment (b, bottom), entirely from the specular reflections. Note the reconstruction of fine details (c) such as a human figure and trees with fall colors through the window. We use the recovered environment for novel view rendering.