AI Paper Atlas

Research map / Detection and segmentation / CNN features

AABO: Adaptive Anchor Box Optimization for Object Detection via Bayesian Sub-sampling

Wenshuo Ma, Tingzhong Tian, Hang Xu, Yimin Huang, Zhenguo Li

ECCV 2020 · Detection and segmentation › CNN features · approach: Box regression and loss design · 21 citations (OpenAlex)

Read the paper

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
optimize anchor settings (scales, ratios) (ablation)Faster-RCNN w FPN · COCOmAP: 36.4 -> 37.1Improved
optimize RPN head architecture (ablation)Faster-RCNN w FPN · COCOmAP: 36.4 -> 36.3No clear effect
optimize anchors and architecture jointly instead of anchors only (ablation)Faster-RCNN w FPN · COCOmAP: 37.1 -> 37.2No clear effect
replace pre-defined anchors with AABO searched anchorsFaster-RCNN w FPN · COCOmAP: 36.4 -> 38.8Improved
replace pre-defined anchors with AABO searched anchorsFaster-RCNN w FPN · VG1000mAP: 6.5 -> 8.0Improved
replace pre-defined anchors with AABO searched anchorsFaster-RCNN w FPN · ADEmAP: 10.3 -> 11.9Improved
replace default anchors with searched anchorsDCNv2 w x101 · COCO (SOTA detectors)mAP: 43.4 -> 45.8Improved
add Bayesian optimization with Hyperband to random search (ablation)Anchor search (single space) · COCOmAP of optimal anchor: 37.2 -> 37.8Improved
use sub-sampling instead of Hyperband in AABO (ablation)AABO (single space) · COCOmAP of optimal anchor: 37.8 -> 38.3Improved
use feature-map-wise search space instead of single space (ablation)AABO w SS · COCOmAP of optimal anchor: 38.3 -> 38.8Improved
use random search in the feature-map-wise space instead of AABO (ablation)Random Search · COCOmAP: 11.5 vs 38.8Hurt
use K-Means anchors instead of pre-defined anchorsFaster-RCNN w FPN · COCOmAP: 36.4 -> 37.0Improved

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