PerQ improves neural image compression with perceptual enhancement and fast decoding
PerQ: Inverse Generative Modeling for Neural Image Compression via Quantization Error Compensation
Computer Vision and Pattern Recognition
Summary
Image compression tools struggle to balance keeping pictures sharp (pixel fidelity) and making them look good to humans (perceptual quality). The authors propose PerQ, a new approach that enhances images by focusing on the range of compressed data instead of guessing from a broad set of possible images. PerQ adds a special model to an existing compression system to improve how images look without losing technical quality or speed. Their tests show PerQ works well on standard image sets and supports many compression levels with one model.
What this means in practice
- •For mobile app developers: Integrate PerQ to enhance image quality perceptually while keeping compression efficient for apps delivering photos at various bitrates.
- •For video streaming engineers: Use PerQ’s approach to improve perceptual quality in video frame compression without retraining models for different quality settings.
Authors
Tianhao Peng, Ho Man Kwan, Fan Zhang, Shan Liu, Dave Bull
Abstract
Neural image codecs face a fundamental trade-off between pixel fidelity and perceptual quality. Distortion-optimized codecs preserve pixel-wise fidelity but produce blurred reconstructions at low bitrates, whereas generative codecs leverage learned priors to improve perceptual quality but tend to exhibit saturating PSNR at higher bitrates and require separate models for different bitrate targets. We propose PerQ, which reframes perceptual reconstruction as inverse generative modeling with bounded support. Specifically, we define the quantization cell as the set of latents that yield the same transmitted symbols and derive a transport bound showing that reconstructions from this cell remain within a rate-determined radius of the codec output. This motivates performing generative reconstruction within the codec-induced quantization cell, rather than sampling from an unconstrained image prior. PerQ implements this principle by attaching a rate-conditioned flow-matching compensator to a frozen, multi-rate distortion-optimized backbone, enabling perceptual enhancement while preserving the backbone's coding performance. Experiments on Kodak and CLIC2020 show that PerQ achieves perceptual performance comparable to generative codecs such as MS-ILLM in LPIPS and FID, while maintaining competitive PSNR relative to distortion-optimized codecs such as ELIC. A single model checkpoint supports a wide bitrate range with fast decoding, and the compensator provides consistent gains across three different backbones. The implementation is available at https://jasminepp.github.io/PerQ/.