← Back to all news
arXiv Computer Vision By Cheng Yuan, Jiawei Shao, Xuelong Li

Information Capacity of Generative Video Compression: Quantifying the Rate-Compute Exchange at Identical Quality

Read the original on arXiv Computer Vision →

The paper introduces the concept of Information Capacity (IC) to quantify how much bandwidth savings a unit of decoder compute can achieve in generative video compression (GVC). By modeling reconstruction quality as a two‑factor power law in data rate and compute, the authors fit measured DISTS of two GVC decoders with high accuracy and define IC as the negative logarithmic slope along an iso‑quality contour. IC is dimensionless, enabling architecture‑agnostic comparisons and revealing that a 14B decoder trades compute for rate far more efficiently than a 1.3B decoder, with significant variation across datasets.

Machine-generated by The Flow from the publisher's headline and feed description — not written or checked by a human. The full article lives at arXiv Computer Vision.

Related stories

arXiv AI
Sep 18

Perceptual Refinement of an End-to-End Video Streaming Pipeline via Generative AI Layers

The paper introduces PRESLEY, an end‑to‑end video streaming pipeline that uses generative AI layers to selectively degrade and restore less important regions of a frame. By replacing destructive block removal with adaptive in‑place degradation and signaling block strength via a side channel, PRESLEY achieves significant bitrate savings and improved background quality compared to its predecessor and pristine baselines. The authors also analyze the theoretical headroom of this architecture, quantifying remaining cost‑axis headroom and modeling post‑restoration damage to guide future rate‑distortion‑restoration selection.

By Emanuele Artioli, Farzad Tashtarian, Christian Timmerer
More like this →
arXiv Computer Vision
3d ago

Rethinking Generative Image Compression at Extremely Low Bitrates

The paper introduces RAE-CoD, a diffusion-based compression method that operates in a representation autoencoder space to preserve recognizable content even at extremely low bitrates. It addresses the problem of semantic collapse observed in existing codecs when the bitrate approaches zero, showing that reconstruction losses conflict with semantic objectives and that VAE diffusion models lose efficiency in preserving semantics. Experiments on MSCOCO-30K demonstrate that RAE-CoD outperforms competitors, reducing VFM feature MSE and Fréchet Distance ratios by at least 25.7% and 69.1% at 0.001–0.008 bpp while maintaining stable recognizability and quality.

By Tianyu Zhang, Zhaoyang Jia, Houqiang Li, Dong Liu
llmsdiffusionmultimodalsafety
More like this →