Neural Radiance Field-Assisted Static-Scene Video Coding

We investigate how to compress static-scene videos efficiently, where each video captures a static scene from different viewpoints. These videos are an essential component of virtual and augmented reality. Current video coding standards like H.265/HEVC have limited compression efficiency due to the limited number of reference frames available, which prevents the full utilization of the abundant temporal redundancy in static-scene videos. To address this limitation, we propose a compression scheme that uses Neural Radiance Fields (NeRF). In our proposed scheme, the encoder derives and encodes the camera parameters, and compresses selected keyframes. The decoder uses an efficient NeRF algorithm to construct an implicit scene model for reconstructing all frames. The experimental results show that our proposed scheme achieves both objective and subjective quality improvement compared to H.265/HEVC, especially at low bit rates.