Implicit Neural Video Representation via Multi-frame Joint Coding

With the rapid development of implicit neural representation (INR) in computer vision, graphics, physical simulation and other fields, it provides new possibilities for processing complex data (such as images, 3D shapes, videos, etc.). However, although implicit neural representation has many advantages, it also has some limitations. The existing INR-based methods are limited to encoding a certain frame of the video separately, which makes it impossible to fully explore the spatiotemporal correlation between adjacent frames in the video, resulting in a lot of redundant information and limited compression efficiency. In addition, multiple frames of data are not jointly analyzed and encoded. There is a large amount of repeated or similar information in the video, which reduces the coding efficiency. In this paper, an implicit neural video representation method based on multi-frame joint encoding (abbreviated as JNeRV) is proposed. In order to evaluate the effectiveness of the proposed model, JNeRV is compared with t he existing implicit neural representation methods NeRV and HNeRV on the Bunny and UVG datasets. The experimental results show that JNeRV has significant improvements over the existing INR baseline methods.