AI Paper Atlas

Research map / Detection and segmentation / CNN features

You Should Look at All Objects

Zhenchao Jin, Dongdong Yu, Luchuan Song, Zehuan Yuan, Lequan Yu

ECCV 2022 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 38 citations (OpenAlex)

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 changedWhereMeasuredEffect
add auxiliary losses to FPN-free Faster R-CNN (ablation)Faster R-CNN ResNet-101 (no FPN) · COCO val2017AP: from 39.0% to 39.6%Improved
add auxiliary losses to FPN-based Faster R-CNN (ablation)Faster R-CNN ResNet-101-FPN · COCO val2017AP: from 39.5% to 39.5%No clear effect
add auxiliary losses (summed) (ablation)RetinaNet ResNet-101-FPN · COCO val2017AP_L: 50.4% to 51.2%Improved
add auxiliary losses (summed) (ablation)Faster R-CNN ResNet-101-FPN · COCO val2017AP_L: 51.6% to 52.2%Improved
add auxiliary losses (summed) (ablation)RetinaNet ResNet-101-FPN · COCO val2017AP_S: 21.7 -> 21.2Hurt
add uncertainty weighting to auxiliary losses (ablation)RetinaNet ResNet-101-FPN · COCO val2017AP_S: 21.2% to 23.3%Improved
add uncertainty weighting to auxiliary losses (ablation)Faster R-CNN ResNet-101-FPN · COCO val2017AP_S: 21.8% to 24.2%Improved
add uncertainty weighting to auxiliary losses (ablation)RetinaNet ResNet-101-FPN · COCO val2017AP: 38.7 -> 40.1Improved
add feature grouping to FPN (ablation)RetinaNet ResNet-101 · COCO val2017AP: 38.5 -> 40.2 (+1.7%)Improved
add feature grouping to FPN (ablation)Faster R-CNN ResNet-101 · COCO val2017AP: 39.5 -> 40.6 (+1.1%)Improved
increase cascade times of feature grouping from 1x to 3x (ablation)RetinaNet ResNet-101 · COCO val2017AP: 40.2 -> 40.8 -> 41.2Improved
increase cascade times of feature grouping from 1x to 3x (ablation)Faster R-CNN ResNet-101 · COCO val2017AP: 40.6 -> 41.7 -> 42.2Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 5 of 5 on most shared results.

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

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