Low-Bitrate Video Compression through Semantic-Conditioned Diffusion

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wang, Lingdong, Su, Guan-Ming, Kothandaraman, Divya, Huang, Tsung-Wei, Hajiesmaili, Mohammad, Sitaraman, Ramesh K.
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913006416297984
author Wang, Lingdong
Su, Guan-Ming
Kothandaraman, Divya
Huang, Tsung-Wei
Hajiesmaili, Mohammad
Sitaraman, Ramesh K.
author_facet Wang, Lingdong
Su, Guan-Ming
Kothandaraman, Divya
Huang, Tsung-Wei
Hajiesmaili, Mohammad
Sitaraman, Ramesh K.
contents Traditional video codecs optimized for pixel fidelity collapse at ultra-low bitrates and produce severe artifacts. This failure arises from a fundamental misalignment between pixel accuracy and human perception. We propose a semantic video compression framework named DiSCo that transmits only the most meaningful information while relying on generative priors for detail synthesis. The source video is decomposed into three compact modalities: a textual description, a spatiotemporally degraded video, and optional sketches or poses that respectively capture semantic, appearance, and motion cues. A conditional video diffusion model then reconstructs high-quality, temporally coherent videos from these compact representations. Temporal forward filling, token interleaving, and modality-specific codecs are proposed to improve multimodal generation and modality compactness. Experiments show that our method outperforms baseline semantic and traditional codecs by 2-10X on perceptual metrics at low bitrates.
format Preprint
id arxiv_https___arxiv_org_abs_2512_00408
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Low-Bitrate Video Compression through Semantic-Conditioned Diffusion
Wang, Lingdong
Su, Guan-Ming
Kothandaraman, Divya
Huang, Tsung-Wei
Hajiesmaili, Mohammad
Sitaraman, Ramesh K.
Computer Vision and Pattern Recognition
Artificial Intelligence
Traditional video codecs optimized for pixel fidelity collapse at ultra-low bitrates and produce severe artifacts. This failure arises from a fundamental misalignment between pixel accuracy and human perception. We propose a semantic video compression framework named DiSCo that transmits only the most meaningful information while relying on generative priors for detail synthesis. The source video is decomposed into three compact modalities: a textual description, a spatiotemporally degraded video, and optional sketches or poses that respectively capture semantic, appearance, and motion cues. A conditional video diffusion model then reconstructs high-quality, temporally coherent videos from these compact representations. Temporal forward filling, token interleaving, and modality-specific codecs are proposed to improve multimodal generation and modality compactness. Experiments show that our method outperforms baseline semantic and traditional codecs by 2-10X on perceptual metrics at low bitrates.
title Low-Bitrate Video Compression through Semantic-Conditioned Diffusion
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2512.00408