Novel View Synthesis for Sparse RGB-D Camera Networks

This paper presents a novel method for view synthesis from sparse neighboring views based on RGB-D images. Distortions and misalignment can occur in the warped images from neighboring viewpoints to the target viewpoint due to errors in the estimated depth map and camera calibration. This degrades the quality of the novel view images, especially when the cameras are far apart. We propose a novel network to estimate the offsets needed to compensate for the misalignment. The experimental results on the ScanNet dataset show that the proposed method outperforms the others. The experimental results on the \textit{ScanNet} dataset show that the proposed method outperforms the state-of-the-art work.