cs.CVNov 8, 2026

Source-Anchored Visual Autoregression for Extreme Image Compression

Authors: Bingxin Wang, Yuan Lan, Yang Xiang, Yiqing Huang

Organizations: Department of Mathematics, HKUST · Huawei Hong Kong Research Center · Tsinghua University

Abstract

In the extreme-low bitrate range below 0.010.01 bpp, image compression becomes a constrained generation problem: the reconstruction depends on inferring plausible missing details while staying close to the input image. Autoregressive token models provide an efficient coarse-to-fine prior for this setting, but the transmitted early-scale tokens often leave object identity and texture underdetermined, which can lead to source-inconsistent reconstructions. We introduce Source-Anchored Visual Autoregression (SAVAR), a generative codec built on Visual Autoregressive (VAR) modeling. SAVAR transmits early-scale tokens and a compact caption. The tokens provide image-specific layout information, while the caption supplies global semantics. SAVAR reconstructs a partial VQ-VAE latent from the retained scales, maps it to the start-of-sequence initialization, and uses the same latent to condition the caption embedding before adaptive layer normalization, separating two roles that are typically driven by a shared condition in VAR-style generation. On the Kodak and CLIC2020 datasets in the extreme-low range, SAVAR reduces perceptual distortion relative to matched ARPC retained-scale operating points, lowering LPIPS by about 88--9%9\% and DISTS by up to 22%22\% at a lower bpp. It reconstructs a 512×512512 \times 512 image in about 218218 ms on one NVIDIA L20X GPU, retaining fast autoregressive decoding in the evaluated setting.

Figures & tables

Appendix figures & tables9 assets

Supplementary material from the paper’s appendix.

Appendix

Explore similar work

CardsList
  1. ResARC: Residual-Aware AutoRegressive Coding for Ultra-Low Bitrate Image Compression

    Sep 30, 2026Qin Yan, Ruixiao Dong, Yutao Xie +5Learned Image CompressionResidual Learning

  2. Rethinking Generative Image Compression at Extremely Low Bitrates

    Sep 30, 2026Tianyu Zhang, Zhaoyang Jia, Houqiang Li +1Learned Image CompressionAutoencoders

  3. Knowledge Distillation for Visual Autoregressive Models

    Jun 4, 2026Elia Peruzzo, Aritra Bhowmik, Guillaume Sautiere +2Visual Autoregressive ModelsAutoregressive Image Generation