Conceptual
Browse Instruct Curate

Layered Motion-Compensated Implicit Neural Representation for Video Compression

A pixel-wise implicit neural representation that compresses video by mapping spatial-temporal coordinates to pixel colors, decomposing the sequence into visually consistent layers each served by a motion-compensating network and an appearance network. Exploiting per-layer motion trajectories removes temporal redundancy, achieving strong rate-distortion while incidentally producing unsupervised segmentation, upsampling, and restoration.