Self-Supervised Visibility Learning for Novel View Synthesis

Figure 1 : Given a few sparse and unstructured input multi-view images, our goal is to synthesize a novel view from a given target camera pose. Our method estimates target-view depth and source-view visibility in an end-to-end self-supervised manner. Compared with the previous state-of-the-art, such as Choi et al. [4] and Riegler and Koltun [24], our method produces superior novel view images of higher quality and with finer details, better conform to the ground-truth.