Research map / Graph learning / Knowledge and temporal graphs
Not All Relations Are Equal: Mining Informative Labels for Scene Graph Generation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 changed | Where | Measured | Effect |
|---|---|---|---|
| Add informative-label imputation and refinement to VCTree-TDE | VCTree-TDE · Visual Genome PredCls, SGCls, and SGDet | mR@20/50/100: PredCls: 16.3/22.9/26.3 -> 22.2/28.1/30.6; SGCls: 11.9/15.8/18.0 -> 17.8/22.0/23.6; SGDet: 6.6/9.0/10.8 -> 8.4/10.3/11.5 | Helped |
| Add informative-label imputation and refinement to VCTree-TDE | VCTree-TDE · Visual Genome zero-shot PredCls | zsR@20/50: 8.98/14.52 -> 9.11/13.52 | Mixed |
| Add informative-label imputation and refinement to VCTree-TDE | VCTree-TDE · Visual Genome zero-shot SGCls and SGDet | zsR@20/50: SGCls: 3.16/4.97 -> 4.26/6.20; SGDet: 1.47/2.3 -> 2.24/3.25 | Helped |
| Train only on implicit rather than all relation labels (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 18.24/24.93 vs 17.85/24.75 | Helped |
| Train only on explicit rather than all relation labels (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 14.06/20.34 vs 17.85/24.75 | Hurt |
| Train on a random relation subset rather than all labels (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 16.99/23.33 vs 17.85/24.75 | Hurt |
| Impute implicit labels after training on explicit labels (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 14.23/20.51 vs 17.85/24.75 | Hurt |
| Impute random labels from a random training subset (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 17.11/23.56 vs 17.85/24.75 | Hurt |
| Add implicit-label imputation to implicit-only training (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 18.24/24.93 -> 21.26/27.14 | Helped |
| Add manifold mixup to baseline training without imputed labels (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 17.85/24.75 -> 17.43/24.20 | Hurt |
| Add manifold mixup to hard imputed labels and refinement (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 20.81/26.78 -> 21.26/27.14 | Helped |
| Remove label refinement while retaining hard imputed labels and mixup (ablation) | Motif-TDE-Sum · Visual Genome PredCls | mR@20/50: 21.26/27.14 -> 19.90/26.35 | Hurt |
Closest papers by meaning
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.91)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 2024 · similarity 0.91)
- Semi-Supervised Clustering Framework for Fine-grained Scene Graph Generation (AAAI 2025 · similarity 0.91)
- Scene Graph-Grounded Image Generation (AAAI 2025 · similarity 0.90)
- RA-SGG: Retrieval-Augmented Scene Graph Generation Framework via Multi-Prototype Learning (AAAI 2025 · similarity 0.90)
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 2024 · similarity 0.90)
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.