AI Paper Atlas

Research map / Detection and segmentation / CNN features

Neural Auto-Exposure for High-Dynamic Range Object Detection

Emmanuel Onzon, Fahim Mannan, Felix Heide

CVPR 2021 · Detection and segmentation › CNN features · approach: Efficient DETR query design · 50 citations (OpenAlex)

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 changedWhereMeasuredEffect
jointly use histogram statistics and detector semantic features for exposure controlNeural auto-exposure for object detection · Synthetic exposure shift k = 1.5, k = 4, and k = 10mAP: 34.01; 34.08; 33.35Improved
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 = 4mAP: 34.01 -> 33.27; 34.08 -> 33.05Hurt
use the histogram branch alone instead of the hybrid model (ablation)Neural auto-exposure for object detection · Synthetic exposure shift k = 10mAP: 33.35 -> 32.17Hurt
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 = 4mAP: 34.01 -> 32.83; 34.08 -> 32.92Hurt
replace the hybrid neural exposure control with average-based AEObject detection pipeline · Experimental vehiclemAP: 32.37 -> 28.80Hurt
replace the hybrid neural exposure control with the strongest conventional baselineObject detection pipeline · Synthetic exposure shift k = 1.5mAP: 34.01 -> 31.73Hurt
replace the hybrid neural exposure control with the strongest conventional baselineObject detection pipeline · Synthetic exposure shift k = 4mAP: 34.08 -> 29.96Hurt
replace the hybrid neural exposure control with the strongest conventional baselineObject detection pipeline · Synthetic exposure shift k = 10mAP: 33.35 -> 26.77Hurt
fine-tune the detector and exposure baseline on HDR data rather than pretraining only on LDR dataGRADIENT AE · Synthetic exposure shift k = 1.5, k = 4, and k = 10mAP: 19.73 -> 31.66; 17.71 -> 27.13; 13.36 -> 20.91Improved
fine-tune the detector and exposure baseline on HDR data rather than pretraining only on LDR dataAVERAGE AE · Synthetic exposure shift k = 1.5, k = 4, and k = 10mAP: 18.94 -> 31.10; 18.02 -> 29.41; 15.35 -> 25.37Improved
include AE in the HDR fine-tuning pipeline rather than fine-tuning without AEGRADIENT AE · Synthetic exposure shift k = 1.5, k = 4, and k = 10mAP: 31.66 -> 31.73; 27.13 -> 28.92; 20.91 -> 22.91Mixed
include AE in the HDR fine-tuning pipeline rather than fine-tuning without AEAVERAGE AE · Synthetic exposure shift k = 1.5, k = 4, and k = 10mAP: 31.10 -> 30.90; 29.41 -> 29.96; 25.37 -> 26.77Mixed

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy