Research map / Image generation / Image restoration
Breaking the Multi-Enhancement Bottleneck: Domain-Consistent Quality Enhancement for Compressed Images
Qunliang Xing, Ce Zheng, Mai Xu, Jing Yang, Shengxi Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| adapt MPRNet with domain-consistent idempotent training | MPRNet · DIV2K, BPG | DI (PSNR, SSIM, TOPIQ, LPIPS): 59% PSNR, 46% SSIM, 52% TOPIQ, 74% LPIPS reduction | Improved |
| adapt CBDNet with domain-consistent training | CBDNet · DIV2K, BPG | DI of PSNR (%): 3.25 -> 0.35 | Improved |
| adapt RBQE with domain-consistent training | RBQE · DIV2K, BPG | DI of PSNR (%): 3.50 -> 1.17 | Improved |
| adapt AR-CNN with domain-consistent training | AR-CNN · DIV2K, BPG | DI of PSNR (%): 1.85 -> 1.75 | Improved |
| adapt Real-ESRGAN with domain-consistent training | Real-ESRGAN · DIV2K, BPG | DI of PSNR (%): 6.88 -> 3.97 | Improved |
| adapt ESRGAN with domain-consistent training | ESRGAN · DIV2K, JPEG | DI of PSNR (%): 7.04 -> 4.97 | Improved |
| adapt RBQE with domain-consistent training | RBQE · DIV2K, JPEG | DI of PSNR (%): 3.51 -> 0.12 | Improved |
| adapt methods with domain-consistent training | all baselines · DIV2K and Flickr2K, single enhancement | average percentage difference vs baseline: 0.197 PSNR, 0.145 SSIM, 1.445 LPIPS, 0.177 TOPIQ | No clear effect |
| supervise multi-enhanced outputs with the raw image (straightforward solution) for CBDNet (ablation) | CBDNet · DIV2K BPG QP37 | once-enhanced PSNR (dB): 31.66 -> 31.56 | Hurt |
| supervise multi-enhanced outputs with the raw image (straightforward solution) for CBDNet (ablation) | CBDNet · DIV2K BPG QP37 | DI of PSNR after fifth enhancement (%): 3.19 -> 2.88 | Improved |
| supervise multi-enhanced outputs with the raw image (straightforward solution) for SRGAN (ablation) | SRGAN · DIV2K BPG QP37 | DI of LPIPS after fifth enhancement (%): 79.22 -> 86.24 | Hurt |
| supervise multi-enhanced outputs with the raw image (straightforward solution) for SRGAN (ablation) | SRGAN · DIV2K BPG QP37 | DI of PSNR after fifth enhancement (%): 8.76 -> 8.72 | No clear effect |
Closest papers by meaning
- Enhancing Quality of Compressed Images by Mitigating Enhancement Bias Towards Compression Domain (CVPR 2024 · similarity 0.92)
- Make Lossy Compression Meaningful for Low-Light Images (AAAI 2024 · similarity 0.86)
- Bridging the Synthetic-to-Authentic Gap: Distortion-Guided Unsupervised Domain Adaptation for Blind Image Quality Assessment (CVPR 2024 · similarity 0.85)
- Multi-Domain Multi-Scale Diffusion Model for Low-Light Image Enhancement (AAAI 2024 · similarity 0.85)
- Denoising as Adaptation: Noise-Space Domain Adaptation for Image Restoration (ICLR 2025 · similarity 0.85)
- UARE: A Unified Vision-Language Model for Image Quality Assessment, Restoration, and Enhancement (CVPR 2026 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
