# Versatile Recompression-Aware Perceptual Image Super-Resolution

Source: [Versatile Recompression-Aware Perceptual Image Super-Resolution](<https://www.sota2.com/research/paper/6925385fd6b0dd334292f774>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-03-31T05:00:13.307Z

Authors: Mingwei He, Tongda Xu, Xingtong Ge, Ming Sun, Chao Zhou, Yan Wang

Published: 2025-11-22T15:12:25.000Z

## Abstract

Perceptual image super-resolution (SR) methods restore degraded images and produce sharp outputs. In practice, those outputs are usually recompressed for storage and transmission. Ignoring recompression is suboptimal as the downstream codec might add additional artifacts to restored images. However, jointly optimizing SR and recompression is challenging, as the codecs are not differentiable and vary in configuration. In this paper, we present \\textbf{Versatile Recompression-Aware Perceptual Super-Resolution (VRPSR)}, which makes existing perceptual SR aware of versatile compression. First, we formulate compression as conditional text-to-image generation and utilize a pre-trained diffusion model to build a generalizable codec simulator. Next, we propose a set of training techniques tailored for perceptual SR, including optimizing the simulator using perceptual targets and adopting slightly compressed images as the training target. Empirically, our VRPSR achieves 10% - 40% bitrate savings based on Real-ESRGAN and S3Diff under H.264/H.265/H.266 single-picture (intra) compression. Besides, our VRPSR facilitates joint optimization of SR and the post-processing model after recompression.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [Perceptual Image Super-Resolution on ImageNet H.266 vvenc (val)](<https://www.sota2.com/research/sota/perceptual-image-super-resolution-on-imagenet-h-266-vvenc-val.md>): Perceptual Image Super-Resolution; ImageNet H.266 vvenc (val); PSNR: 42.02.

- [Perceptual Image Super-Resolution on ImageNet H.264 x264 (val)](<https://www.sota2.com/research/sota/perceptual-image-super-resolution-on-imagenet-h-264-x264-val.md>): Perceptual Image Super-Resolution; ImageNet H.264 x264 (val); PSNR: 28.5.

- [Image Super-Resolution on Kodak H.264 x264](<https://www.sota2.com/research/sota/image-super-resolution-on-kodak-h-264-x264.md>): Image Super-resolution; Kodak H.264 x264; PSNR: 27.64.

- [Image Super-Resolution on Kodak H.265 x265](<https://www.sota2.com/research/sota/image-super-resolution-on-kodak-h-265-x265.md>): Image Super-resolution; Kodak H.265 x265; PSNR: 27.71.

- [Image Super-Resolution on Kodak H.266 vvenc](<https://www.sota2.com/research/sota/image-super-resolution-on-kodak-h-266-vvenc.md>): Image Super-resolution; Kodak H.266 vvenc; PSNR: 27.77.

- [Perceptual Image Super-Resolution on ImageNet H.265 x265 (val)](<https://www.sota2.com/research/sota/perceptual-image-super-resolution-on-imagenet-h-265-x265-val.md>): Perceptual Image Super-Resolution; ImageNet H.265 x265 (val); PSNR: 28.64.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2511.18090>)

[PDF](<https://arxiv.org/pdf/2511.18090.pdf>)