HiFiPix improves real-world image super-resolution with high-detail pixel restoration
HiFiPix: Bringing High-Fidelity Real-World Super-Resolution Back to Pixel Space
Computer Vision and Pattern Recognition
Summary
Making blurry or low-resolution photos clearer is a common challenge. The authors found that popular methods that compress images internally lose important details. They created HiFiPix, a system that works directly with the full image to better keep fine details and avoid artifacts. By using special tools to prevent patchy edges and carefully tuning parts of the system, HiFiPix restores images more accurately and beautifully. Their tests show it works better than earlier approaches at both accuracy and visual quality.
What this means in practice
- •For mobile app developers: Integrate HiFiPix to enhance photo quality within image editing apps by producing sharper and more natural super-resolved images from real-world photos.
- •For graphic design studios: Use HiFiPix to achieve ultra high-fidelity image enlargement without artifacts for print and digital media projects requiring precise detail restoration.
Authors
Wenjie Li, Du Chen, Hui Wang, Linlong Fan, Youwei Lv, Zeyi Chen, Zehan Chen, Jinwei Chen, Heng Guo, Zhanyu Ma, Qingnan Fan
Abstract
Diffusion-based real-world image super-resolution suffers from limited reconstruction fidelity due to information loss in VAE-based latent representations. Recently, pixel-space diffusion models offer a promising high-faithfulness alternative by avoiding latent compression, but adapting them to Real-ISR remains challenging due to inherent patch boundary artifacts and degraded generative capability. To address the issues and maximize the potential of pretrained pixel-space diffusion models, we propose HiFiPix, a pixel-space Real-ISR framework capable of ultra high-fidelity reconstruction and fine-grained visual generation effects. To suppress patch boundary artifacts, we penalize unrealistic high-frequency periodic components with a Cross-Scale Wavelet Discriminator (CWSD) for image-level supervision, and Mixup Contrastive Self-Distillation (MixCSD) for intermediate representation regularization. Furthermore, we uncover a heterogeneous generation-fidelity trade-off across low-rank adapters (LoRA) and propose Selective LoRA Pruning (LoRA-SP) to adaptively prune LoRA layers, enhancing generative capability with minimal deterioration of reconstruction fidelity. Extensive experiments demonstrate that HiFiPix consistently outperforms existing state-of-the-art latent space methods in terms of both restoration accuracy and perceptual quality by a large margin. Code and models will be released.