AI Paper Atlas

Research map / Detection and segmentation / CNN features

DeFRCN: Decoupled Faster R-CNN for Few-Shot Object Detection

Limeng Qiao, Yuxuan Zhao, Zhiyuan Li, Xi Qiu, Jianan Wu, Chi Zhang

ICCV 2021 · Detection and segmentation › CNN features · approach: Region-based and query-based detectors · 445 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 GDL in novel fine-tuning (GDL-N) (ablation)Faster R-CNN · COCO novel FSOD, 10-shotmAP: 7.9 -> 15.2Improved
add GDL in novel fine-tuning (GDL-N) (ablation)Faster R-CNN · COCO novel FSOD, 30-shotmAP: 12.2 -> 19.0Improved
add GDL in base training (GDL-B) (ablation)Faster R-CNN · COCO novel FSOD, 10-shotmAP: 7.9 -> 8.2Improved
add GDL in base training (GDL-B) on top of GDL-N (ablation)Faster R-CNN + GDL-N · COCO novel FSOD, 10-shotmAP: 15.2 -> 16.9Improved
add Prototypical Calibration Block (ablation)Faster R-CNN · COCO novel FSOD, 10-shotmAP: 7.9 -> 10.4Improved
add PCB on top of GDL-B and GDL-N (ablation)DeFRCN · COCO novel FSOD, 10-shotmAP: 16.9 -> 18.5Improved
add PCB to TFA (ablation)TFA · COCO novel FSOD, 10-shotmAP: 9.0 -> 11.8Improved
add PCB to MPSR (ablation)MPSR · COCO novel FSOD, 30-shotmAP: 14.5 -> 15.5Improved
add PCB to FRCN-ft (ablation)FRCN-ft · COCO novel FSOD, 10-shotmAP: 6.9 -> 9.9Improved
use stronger pre-trained model IG-WSL instead of IN-1K in PCB (ablation)DeFRCN · VOC Novel Set 1, 10-shotAP50: 60.8 -> 68.2Improved
use self-supervised IN-SwAV instead of IN-1K in PCB (ablation)DeFRCN · VOC Novel Set 1, 10-shotAP50: 60.8 -> 56.3Hurt
add GDL to conventional Faster R-CNN (ablation)R50-C4-1x · COCO conventional detectionAP50: 56.1 -> 57.6Improved

Later papers that compared with this method

21 later accepted papers report DeFRCN as a baseline in their results tables; 20 of them report a better number on most of the results they share with it.

13 more not listed.

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