AI Paper Atlas

Research map / Graph learning / Knowledge and temporal graphs

From General to Specific: Informative Scene Graph Generation via Balance Adjustment

Yuyu Guo, Lianli Gao, Xuanhan Wang, Yuxuan Hu, Xing Xu, Xu Lu, Heng Tao Shen, Jingkuan Song

ICCV 2021 · Graph learning › Knowledge and temporal graphs · approach: Scene graph generation · 98 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 Balanced Predicate Learning (ablation)Transformer · Visual Genome PredClsmR@100: 17.5 -> 31.7Improved
add Balanced Predicate Learning (ablation)MotifNet · Visual Genome PredClsmR@100: 15.8 -> 29.1Improved
add Balanced Predicate Learning (ablation)VCTree · Visual Genome SGDetmR@100: 6.9 -> 14.4Improved
add Semantic Adjustment on top of BPL (ablation)Transformer+BPL · Visual Genome PredClsmR@100: 31.7 -> 34.2Improved
add Semantic Adjustment on top of BPL (ablation)VCTree+BPL · Visual Genome SGClsmR@100: 19.5 -> 21.2Improved
use frozen SA matrix instead of finetuning it (ablation)Transformer+BPL+SA · Visual Genome PredClsmR@100: 32.1 -> 34.2Improved
use SA matrix from confusion statistics instead of random init (ablation)Transformer+BPL · Visual Genome PredClsmR@100: 31.3 -> 32.1Improved
add identity matrix weight alpha=1.0 instead of 0.0 (ablation)SA transition matrix · Visual Genome PredClsmR@100: 27.3 -> 34.2Improved
build target domain from Visual Genome instead of Wikipedia (ablation)BPL · Visual Genome PredClsmR@100: 28.0 -> 34.2Improved
train on Wikipedia-based target domain instead of general domain (ablation)Transformer · Visual Genome PredClsmR@100: 17.5 -> 28.0Improved
use pretrained-model undersampling instead of random (ablation)Transformer+BPL+SA · Visual Genome PredClsmR@20: 26.7 -> 28.5Improved
use pretrained-model undersampling instead of random (ablation)Transformer+BPL+SA · Visual Genome SGDetmR@20: 11.4 -> 11.4No clear effect

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