AI Paper Atlas

Research map / Detection and segmentation / Open-set and few-shot detection

Query Adaptive Few-Shot Object Detection With Heterogeneous Graph Convolutional Networks

Guangxing Han, Yicheng He, Shiyuan Huang, Jiawei Ma, Shih-Fu Chang

ICCV 2021 · Detection and segmentation › Open-set and few-shot detection · approach: Few-shot object detection · 158 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 heterogeneous GCNs (full model) instead of none (ablation)QA-FewDet · MSCOCO 2/10/30-shot (meta-training)AP: 5.4 -> 7.8 (2-shot), 7.6 -> 10.2 (10-shot), 8.9 -> 11.5 (30-shot)Improved
replace MLP with GCN message passing (ablation)QA-FewDet · MSCOCO 10-shot (meta-training)AP: 8.4 -> 10.2Improved
add class-proposal edges to proposal-proposal edges (ablation)QA-FewDet · MSCOCO 2-shot (meta-training)AP: 6.7 -> 7.6Improved
add proposal-proposal edges to class-proposal edges (ablation)QA-FewDet · MSCOCO 10-shot (meta-training)AP: 9.8 -> 10.0Improved
add Inter-Class Subgraph (class-class edges) (ablation)QA-FewDet · MSCOCO 10-shot (meta-training)AP: 10.0 -> 10.2Improved
fine-tune on novel classes after meta-trainingQA-FewDet · MSCOCO 30-shotAP: 11.5 -> 16.5Improved
fine-tune on novel classes after meta-trainingQA-FewDet · MSCOCO 2-shotAP: 7.8 -> 7.6No clear effect
use bidirectional class-proposal edge instead of class->proposal only (ablation)QA-FewDet · MSCOCO 10-shot (meta-training)AP: 8.6 -> 9.8Improved
use class-proposal edge instead of non-local attention (ablation)QA-FewDet · MSCOCO 10-shot (meta-training)AP: 8.3 -> 9.8Improved
add local and global context to proposal-proposal edge (ablation)QA-FewDet · MSCOCO 10-shot (meta-training)AP: 9.0 -> 9.3Improved
add all 60 base classes to Inter-Class Subgraph (ablation)QA-FewDet · MSCOCO 10-shot (meta-training)AP: 8.5 -> 9.0Improved
learn W in Inter-Class Subgraph GCN layer (ablation)QA-FewDet · MSCOCO 10-shot (meta-training)AP: 9.0 -> 7.2Hurt

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