Research map / Vision-language and multimodal / Human and embodied modeling
RAW-Adapter: Adapting Pretrained Visual Model to Camera RAW Images
Ziteng Cui, Tatsuya Harada
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 kernel predictor P_K (gain ratio and adaptive Gaussian denoise) (ablation) | RAW-Adapter on RetinaNet ResNet-50 · PASCAL RAW (dark) | mAP: 82.6 -> 85.0 | Improved |
| add kernel predictor P_K (ablation) | RAW-Adapter on RetinaNet ResNet-50 · PASCAL RAW (normal) | mAP: 89.2 -> 89.2 | No clear effect |
| add matrix predictor P_M (white balance and CCM) (ablation) | RAW-Adapter · PASCAL RAW (dark) | mAP: 85.0 -> 86.2 | Improved |
| add implicit 3D LUT (ablation) | RAW-Adapter · PASCAL RAW (over-exp and dark) | mAP: dark 86.2 -> 86.3; normal 89.5 -> 89.4 | No clear effect |
| add model-level adapters M (ablation) | RAW-Adapter · PASCAL RAW (normal/over-exp/dark) | mAP: dark 86.3 -> 86.6; normal 89.4 -> 89.7; over-exp 89.0 -> 89.5 | Improved |
| add model-level adapters M (ablation) | RAW-Adapter on LOD · LOD, RetinaNet | mAP: 61.6 -> 62.1 | Improved |
| increase NILUT MLP dim from 32 to 64 or 128 (ablation) | RAW-Adapter · ADE20K RAW | mIoU: 47.95 -> 48.02 (dim 64) | No clear effect |
| reduce NILUT MLP dim from 32 to 16 (ablation) | RAW-Adapter · ADE20K RAW (normal) | mIoU: 47.95 -> 47.37 | Hurt |
| apply human-vision ISP methods (InvISP, LiteISP) to RAW before segmentation in dark | Segformer MIT-B5 · ADE20K RAW (dark) | mIoU: 37.55 (demosaicing) -> 4.03 (InvISP) | Hurt |
| use RAW-Adapter instead of Dirty-Pixel | RetinaNet ResNet-50 · PASCAL RAW (dark) | mAP: 83.6 -> 86.6 | Improved |
| use RAW-Adapter on small backbone MIT-B0 | Segformer · ADE20K RAW (dark) | mIoU: 24.89 -> 25.06 (vs Dirty-Pixel) | Improved |
Closest papers by meaning
- RAW-Adapter: Adapting Pre-Trained Visual Model to Camera RAW Images and a Benchmark (TPAMI 2026 · similarity 0.96)
- Rawformer: Unpaired Raw-to-Raw Translation for Learnable Camera ISPs (ECCV 2024 · similarity 0.88)
- Model-Based Image Signal Processors via Learnable Dictionaries (AAAI 2022 · similarity 0.86)
- Beyond RGB: Adaptive Parallel Processing for RAW Object Detection (ICCV 2025 · similarity 0.86)
- ReRAW: RGB-to-RAW Image Reconstruction via Stratified Sampling for Efficient Object Detection on the Edge (CVPR 2025 · similarity 0.84)
- Leveraging Frame Affinity for sRGB-to-RAW Video De-rendering (CVPR 2024 · similarity 0.83)
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.
