Research map / Detection and segmentation / Open-set and few-shot detection
CLP: A Real-World Dataset of Contaminated Lens Protectors for Robust Semantic Segmentation
Sungyong Park, Sooyoung Choi, Hyunsuh Koh, Youngjae Choi, Heewon Kim
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 |
|---|---|---|---|
| train on clean images only with domain generalization (SoMA, Rein) instead of domain adaptation with unlabeled contaminated images | segmentation models · CLP test | mIoU: +14.0 (SoMA) and +12.9 (Rein) mIoU over best DA baseline | Improved |
| use frozen DINOv2 foundation encoder with Mask2Former decoder | Mask2Former · CLP test overall | mIoU: 36.4 -> 47.2 | Improved |
| use frozen SAM encoder instead of DINOv2 | Mask2Former · CLP test overall | mIoU: 47.2 -> 39.2 | Hurt |
| use RobustSAM encoder instead of SAM | SAM + Mask2Former · CLP mud | mIoU: 30.2 -> 33.6 | Improved |
| apply FIFO fog-invariant training | segmentation · CLP test overall | mIoU: 36.4 -> 32.2 | Hurt |
| add generic pretrained URIE restoration before Mask2Former (ablation) | Mask2Former · CLP test overall | mIoU: 33.9 -> 32.2 | Hurt |
| add DiffUIR restoration (best PSNR) before Mask2Former (ablation) | Mask2Former · CLP test overall | mIoU: 33.9 -> 35.7; water -1.9, mud -0.3, condensation +2.4 | Mixed |
| add task-aware UniRestore before Mask2Former (ablation) | Mask2Former · CLP test overall | mIoU: 33.9 -> 35.5 (+1.6) | Improved |
| add NAFNet restoration before Mask2Former (ablation) | Mask2Former · CLP test overall | mIoU: 33.9 -> 33.7 | Hurt |
| increase training data from 10% to 100% of CLP scenes (ablation) | SoMA · CLP test | mIoU: 22.19 -> 48.69 | Improved |
| increase training data from 10% to 100% of CLP scenes (ablation) | NAFNet · CLP test restoration | PSNR: 24.63 -> 25.77 | Improved |
| increase lens-to-protector distance 0 to 5 cm under mud | Mask2Former · CLP mud | mIoU: 34.5 -> 26.1 | Hurt |
Closest papers by meaning
- Semantic Segmentation on Raindrop Degraded Images Using Two-Stage Dual Teacher-Student Learning (AAAI 2025 · similarity 0.85)
- Reliability in Semantic Segmentation: Can We Use Synthetic Data? (ECCV 2024 · similarity 0.85)
- SIDL: A Real-World Dataset for Restoring Smartphone Images with Dirty Lenses (AAAI 2025 · similarity 0.84)
- v-CLR: View-Consistent Learning for Open-World Instance Segmentation (CVPR 2025 · similarity 0.83)
- LL-Bench: Rethinking Low-Level Vision Evaluation in the Era of Large-Scale Generative Models (NeurIPS 2026 · similarity 0.82)
- Hyperbolic Safety-Aware Vision-Language Models (CVPR 2025 · similarity 0.82)
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.
