Research map / Detection and segmentation / CNN features
FeatEnHancer: Enhancing Hierarchical Features for Object Detection and Beyond Under Low-Light Vision
Khurram Azeem Hashmi, Goutham Kallempudi, Didier Stricker, Muhammad Zeshan Afzal
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 FeatEnHancer to FQ R-CNN instead of LLIE preprocessing | FQ R-CNN · ExDark | mAP50: 74.5 -> 86.3 | Improved |
| add FeatEnHancer to RetinaNet | RetinaNet · ExDark | mAP50: 72.1 -> 72.6 | Improved |
| add FeatEnHancer to RetinaNet | RetinaNet · DARK FACE | AP50: 47.3 -> 47.2 | No clear effect |
| add FeatEnHancer to FQ R-CNN | FQ R-CNN · DARK FACE | AP50: 67.5 -> 69.0 | Improved |
| add FeatEnHancer to DeepLabV3+ | DeepLabV3+ · ACDC nighttime | mIoU: 45.7 -> 54.9 | Improved |
| preprocess frames with existing LLIE methods | SELSA · DarkVision | mAP: 32.8 -> 7.83 (Zero-DCE, illumination 3.2) | Hurt |
| add FeatEnHancer to SELSA | SELSA · DarkVision illumination 3.2 | mAP: 32.8 -> 34.6 | Improved |
| use scale-aware attentional feature aggregation (SAFA) instead of averaging or skip connections (ablation) | FeatEnHancer · ExDark | mAP: 70.3 -> 72.6 | Improved |
| apply SAFA first then skip connection rather than reversed order (ablation) | FeatEnHancer · ExDark | mAP: 70.9 -> 72.6 | Improved |
| use convolutional downsampling instead of pooling or interpolation (ablation) | FeatEnHancer · ExDark | mAP: 70.7 -> 72.6 | Improved |
| use scales (4, 8) instead of (8, 16) (ablation) | FeatEnHancer · ACDC | mIoU: 45.6 -> 54.9 | Improved |
| increase attention blocks N from 8 to 12 (ablation) | SAFA · ExDark | mAP: 72.6 -> 72.4 | No clear effect |
Later papers that compared with this method
4 later accepted papers report FeatEnHancer as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- TorchAdapt: Towards Light-Agnostic Real-Time Visual Perception (ICCV 2025 · better on 8 of 8 shared results)
- Dark-ISP: Enhancing RAW Image Processing for Low-Light Object Detection (ICCV 2025 · better on 4 of 5 shared results)
- Beyond RGB: Adaptive Parallel Processing for RAW Object Detection (ICCV 2025 · better on 3 of 3 shared results)
- FRBNet: Revisiting Low-Light Vision through Frequency-Domain Radial Basis Network (NeurIPS 2025 · better on 1 of 1 shared result)
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
- Boosting Object Detection with Zero-Shot Day-Night Domain Adaptation (CVPR 2024 · similarity 0.86)
- FRBNet: Revisiting Low-Light Vision through Frequency-Domain Radial Basis Network (NeurIPS 2025 · similarity 0.85)
- Trash to Treasure: Low-Light Object Detection via Decomposition-and-Aggregation (AAAI 2024 · similarity 0.85)
- You Only Look Around: Learning Illumination-Invariant Feature for Low-light Object Detection (NeurIPS 2024 · similarity 0.85)
- A New Multi-Source Light Detection Benchmark and Semi-Supervised Focal Light Detection (NeurIPS 2024 · similarity 0.84)
- NightHaze: Nighttime Image Dehazing via Self-Prior Learning (AAAI 2025 · similarity 0.84)
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.
