Extreme Video Compression with Pre-trained Diffusion Models

TL;DR

This paper introduces a novel extreme video compression method using pre-trained diffusion models at the decoder, achieving high perceptual quality at very low bit rates by sequentially predicting frames and restarting when quality drops.

Contribution

The paper presents a new approach leveraging diffusion-based generative models for low-bpp video compression, demonstrating effective frame prediction and quality maintenance.

Findings

Achieves perceptually high-quality video reconstruction at 0.02 bpp.

Outperforms standard codecs like H.264 and H.265 in low bit rate regimes.

Uses temporal relations in video data with generative models.

Abstract

Diffusion models have achieved remarkable success in generating high quality image and video data. More recently, they have also been used for image compression with high perceptual quality. In this paper, we present a novel approach to extreme video compression leveraging the predictive power of diffusion-based generative models at the decoder. The conditional diffusion model takes several neural compressed frames and generates subsequent frames. When the reconstruction quality drops below the desired level, new frames are encoded to restart prediction. The entire video is sequentially encoded to achieve a visually pleasing reconstruction, considering perceptual quality metrics such as the learned perceptual image patch similarity (LPIPS) and the Frechet video distance (FVD), at bit rates as low as 0.02 bits per pixel (bpp). Experimental results demonstrate the effectiveness of the proposed scheme compared to standard codecs such as H.264 and H.265 in the low bpp regime. The results showcase the potential of exploiting the temporal relations in video data using generative models. Code is available at: https://github.com/ElesionKyrie/Extreme-Video-Compression-With-Prediction-Using-Pre-trainded-Diffusion-Models-