NeuralLVC: Neural Lossless Video Compression via Masked Diffusion with Temporal Conditioning

TL;DR

NeuralLVC introduces a neural lossless video codec combining masked diffusion and temporal conditioning, outperforming traditional codecs in lossless compression of video and images.

Contribution

The paper presents NeuralLVC, a novel neural lossless video compression method using masked diffusion and an I/P-frame architecture for improved redundancy exploitation.

Findings

NeuralLVC outperforms H.264 and H.265 lossless codecs on Xiph CIF sequences.

Exact reconstruction verified through end-to-end encode-decode testing.

Group-wise decoding offers controllable speed and compression trade-offs.

Abstract

While neural lossless image compression has advanced significantly with learned entropy models, lossless video compression remains largely unexplored in the neural setting. We present NeuralLVC, a neural lossless video codec that combines masked diffusion with an I/P-frame architecture for exploiting temporal redundancy. Our I-frame model compresses individual frames using bijective linear tokenization that guarantees exact pixel reconstruction. The P-frame model compresses temporal differences between consecutive frames, conditioned on the previous decoded frame via a lightweight reference embedding that adds only 1.3% trainable parameters. Group-wise decoding enables controllable speed-compression trade-offs. Our codec is lossless in the input domain: for video, it reconstructs YUV420 planes exactly; for image evaluation, RGB channels are reconstructed exactly. Experiments on 9 Xiph CIF sequences show that NeuralLVC outperforms H.264 and H.265 lossless by a significant margin. We verify exact reconstruction through end-to-end encode-decode testing with arithmetic coding. These results suggest that masked diffusion with temporal conditioning is a promising direction for neural lossless video compression.