Research map / Detection and segmentation / Open-set and few-shot detection
Restoration Adaptation for Semantic Segmentation on Low Quality Images
Kai Guan, Rongyuan Wu, Shuai Li, Wentao Zhu, Wenjun Zeng, Lei Zhang
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 |
|---|---|---|---|
| add restoration adaptation with SCR before segmentation training (ablation) | RASS · ADE20K (simulated LQ) | mIoU: 44.56 -> 47.42 | Improved |
| use SCR instead of OSEDiff+ for restoration adaptation (ablation) | RASS · ADE20K (simulated LQ) | mIoU: 47.05 -> 47.42 | Improved |
| use SCR instead of OSEDiff as restoration before segmentation (R2S) (ablation) | RASS · ADE20K (simulated LQ) | mIoU: 46.05 -> 46.81 | Improved |
| add VAE encoder features to UNet features (ablation) | RASS · ADE20K (simulated LQ) | mIoU: 46.85 -> 47.42 | Improved |
| add VAE decoder features to UNet features (ablation) | RASS · ADE20K (simulated LQ) | mIoU: 46.85 -> 47.25 | Improved |
| add both VAE encoder and decoder features (ablation) | RASS · ADE20K (simulated LQ) | mIoU: 47.42 -> 47.60 | Improved |
| add additional descriptions (AD) in SCR training (ablation) | SCR · DrealSR | MANIQA: 0.6338 -> 0.6394 | Improved |
| enable additional descriptions at inference (ablation) | SCR · DrealSR | MUSIQ: 67.15 -> 67.16 | No clear effect |
| reduce SCR training data from all 84,991 to 1,000 LSDIR images (ablation) | SCR · DrealSR | PSNR: 28.54 -> 28.45 | No clear effect |
| use 475 curated LSDIR images instead of all images (ablation) | SCR · DrealSR | MUSIQ: 64.51 -> 69.12; PSNR 28.54 -> 27.43 | Mixed |
| add 518 enhanced ADE20K images to the 475 LSDIR images (ablation) | SCR · DrealSR | PSNR: 27.43 -> 28.14 | Improved |
| use fine-tuning on LQ data instead of direct testing for baselines on RealLQ | FeedFormer and SeaFormer · RealLQ | mIoU: 20.87 -> 19.19; 15.52 -> 14.97 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SegFormer — SegFormer: A Topic Segmentation Model with Controllable Range of Attention (AAAI 2023) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- UARE: A Unified Vision-Language Model for Image Quality Assessment, Restoration, and Enhancement (CVPR 2026 · similarity 0.86)
- Convolution Meets LoRA: Parameter Efficient Finetuning for Segment Anything Model (ICLR 2024 · similarity 0.85)
- Sharing Key Semantics in Transformer Makes Efficient Image Restoration (NeurIPS 2024 · similarity 0.85)
- ReFIR: Grounding Large Restoration Models with Retrieval Augmentation (NeurIPS 2024 · similarity 0.85)
- S2S2: Semantic Stacking for Robust Semantic Segmentation in Medical Imaging (AAAI 2025 · similarity 0.85)
- SemiDAViL: Semi-supervised Domain Adaptation with Vision-Language Guidance for Semantic Segmentation (CVPR 2025 · 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.
