Research map / Detection and segmentation / CNN features
Neural Auto-Exposure for High-Dynamic Range Object Detection
Emmanuel Onzon, Fahim Mannan, Felix Heide
Check 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 |
|---|---|---|---|
| jointly use histogram statistics and detector semantic features for exposure control | Neural auto-exposure for object detection · Synthetic exposure shift k = 1.5, k = 4, and k = 10 | mAP: 34.01; 34.08; 33.35 | Improved |
| use the semantic branch alone instead of the hybrid model (ablation) | Neural auto-exposure for object detection · Synthetic exposure shift k = 1.5 and k = 4 | mAP: 34.01 -> 33.27; 34.08 -> 33.05 | Hurt |
| use the histogram branch alone instead of the hybrid model (ablation) | Neural auto-exposure for object detection · Synthetic exposure shift k = 10 | mAP: 33.35 -> 32.17 | Hurt |
| use the histogram branch alone instead of the hybrid model (ablation) | Neural auto-exposure for object detection · Synthetic exposure shift k = 1.5 and k = 4 | mAP: 34.01 -> 32.83; 34.08 -> 32.92 | Hurt |
| replace the hybrid neural exposure control with average-based AE | Object detection pipeline · Experimental vehicle | mAP: 32.37 -> 28.80 | Hurt |
| replace the hybrid neural exposure control with the strongest conventional baseline | Object detection pipeline · Synthetic exposure shift k = 1.5 | mAP: 34.01 -> 31.73 | Hurt |
| replace the hybrid neural exposure control with the strongest conventional baseline | Object detection pipeline · Synthetic exposure shift k = 4 | mAP: 34.08 -> 29.96 | Hurt |
| replace the hybrid neural exposure control with the strongest conventional baseline | Object detection pipeline · Synthetic exposure shift k = 10 | mAP: 33.35 -> 26.77 | Hurt |
| fine-tune the detector and exposure baseline on HDR data rather than pretraining only on LDR data | GRADIENT AE · Synthetic exposure shift k = 1.5, k = 4, and k = 10 | mAP: 19.73 -> 31.66; 17.71 -> 27.13; 13.36 -> 20.91 | Improved |
| fine-tune the detector and exposure baseline on HDR data rather than pretraining only on LDR data | AVERAGE AE · Synthetic exposure shift k = 1.5, k = 4, and k = 10 | mAP: 18.94 -> 31.10; 18.02 -> 29.41; 15.35 -> 25.37 | Improved |
| include AE in the HDR fine-tuning pipeline rather than fine-tuning without AE | GRADIENT AE · Synthetic exposure shift k = 1.5, k = 4, and k = 10 | mAP: 31.66 -> 31.73; 27.13 -> 28.92; 20.91 -> 22.91 | Mixed |
| include AE in the HDR fine-tuning pipeline rather than fine-tuning without AE | AVERAGE AE · Synthetic exposure shift k = 1.5, k = 4, and k = 10 | mAP: 31.10 -> 30.90; 29.41 -> 29.96; 25.37 -> 26.77 | Mixed |
Closest papers by meaning
- Neural Exposure Fusion for High-Dynamic Range Object Detection (CVPR 2024 · similarity 0.92)
- Real-Time Scene-Adaptive Tone Mapping for High-Dynamic Range Object Detection (NeurIPS 2025 · similarity 0.88)
- Self-Supervised High Dynamic Range Imaging with Multi-Exposure Images in Dynamic Scenes (ICLR 2024 · similarity 0.87)
- High Dynamic Range Novel View Synthesis with Single Exposure (ICML 2025 · similarity 0.85)
- Color Event Enhanced Single-Exposure HDR Imaging (AAAI 2024 · similarity 0.85)
- EvHDR-NeRF: Building High Dynamic Range Radiance Fields with Single Exposure Images and Events (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.
