Conditional Video Generation for High-Efficiency Video Compression

TL;DR

This paper introduces a novel video compression framework using conditional diffusion models that optimize perceptual quality by synthesizing videos from sparse signals, outperforming traditional codecs especially at high compression ratios.

Contribution

It proposes a new approach framing video compression as a conditional generation task with multi-granular conditioning and robust training strategies.

Findings

Outperforms traditional and neural codecs on perceptual metrics

Achieves higher perceptual quality at high compression ratios

Demonstrates robustness through multi-condition training

Abstract

Perceptual studies demonstrate that conditional diffusion models excel at reconstructing video content aligned with human visual perception. Building on this insight, we propose a video compression framework that leverages conditional diffusion models for perceptually optimized reconstruction. Specifically, we reframe video compression as a conditional generation task, where a generative model synthesizes video from sparse, yet informative signals. Our approach introduces three key modules: (1) Multi-granular conditioning that captures both static scene structure and dynamic spatio-temporal cues; (2) Compact representations designed for efficient transmission without sacrificing semantic richness; (3) Multi-condition training with modality dropout and role-aware embeddings, which prevent over-reliance on any single modality and enhance robustness. Extensive experiments show that our method significantly outperforms both traditional and neural codecs on perceptual quality metrics such as Fr\'echet Video Distance (FVD) and LPIPS, especially under high compression ratios.