Research map / Detection and segmentation / CNN features
DarkFeat: Noise-Robust Feature Detector and Descriptor for Extremely Low-Light RAW Images
Yuze He, Yubin Hu, Wang Zhao, Jisheng Li, Yong-Jin Liu, Yuxing Han, Jiangtao Wen
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 selective suppression constraint instead of training noisy-image detection scores without it (ablation) | DarkFeat · MID indoor and outdoor | Nτ@5: 0.461 -> 0.606 (indoor); 0.395 -> 0.497 (outdoor) | Improved |
| Add selective suppression constraint instead of training without it (ablation) | DarkFeat · MID indoor and outdoor | Nτ@10: 0.571 -> 0.689 (indoor); 0.507 -> 0.620 (outdoor) | Improved |
| Replace attention loss with selective suppression constraint (ablation) | DarkFeat · MID indoor and outdoor | Nτ@5: 0.448 -> 0.606 (indoor); 0.374 -> 0.497 (outdoor) | Improved |
| Replace score-map MSE with selective suppression constraint (ablation) | DarkFeat · MID indoor and outdoor | Nτ@5: 0.470 -> 0.606 (indoor); 0.404 -> 0.497 (outdoor) | Improved |
| Replace descriptor MSE with selective suppression constraint (ablation) | DarkFeat · MID indoor and outdoor | Nτ@5: 0.535 -> 0.606 (indoor); 0.463 -> 0.497 (outdoor) | Improved |
| Add noise robustness map instead of using selective suppression alone (ablation) | DarkFeat · MID indoor and outdoor | Nτ@5: 0.606 -> 0.618 (indoor); 0.497 -> 0.519 (outdoor) | Improved |
| Add noise robustness map instead of using selective suppression alone (ablation) | DarkFeat · MID indoor and outdoor | Nτ@10: 0.689 -> 0.705 (indoor); 0.620 -> 0.633 (outdoor) | Improved |
| Use vanilla reliability trained on normal-image AP rather than noise robustness map (ablation) | DarkFeat · MID indoor and outdoor | Nτ@5: 0.618 -> 0.606 (indoor); 0.519 -> 0.493 (outdoor) | Hurt |
| Use vanilla reliability trained on noisy-image AP rather than noise robustness map (ablation) | DarkFeat · MID indoor and outdoor | Nτ@5: 0.618 -> 0.571 (indoor); 0.519 -> 0.486 (outdoor) | Hurt |
| Use DarkFeat directly on RAW rather than add HistEQ preprocessing | DarkFeat · MID indoor and outdoor | Nτ@5: 0.618 -> 0.617 (indoor); 0.519 -> 0.500 (outdoor) | Mixed |
Closest papers by meaning
- Matching in the Dark: A Dataset for Matching Image Pairs of Low-Light Scenes (ICCV 2021 · similarity 0.87)
- FRBNet: Revisiting Low-Light Vision through Frequency-Domain Radial Basis Network (NeurIPS 2025 · similarity 0.85)
- Learning to See in the Dark (CVPR 2018 · similarity 0.85)
- XFeat: Accelerated Features for Lightweight Image Matching (CVPR 2024 · similarity 0.85)
- You Only Look Around: Learning Illumination-Invariant Feature for Low-light Object Detection (NeurIPS 2024 · similarity 0.83)
- Boosting Object Detection with Zero-Shot Day-Night Domain Adaptation (CVPR 2024 · 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.
