Research map / Graph learning / Knowledge and temporal graphs
Semantic Diversity-aware Prototype-based Learning for Unbiased Scene Graph Generation
Jaehyeong Jeon, Kibum Kim, Kanghoon Yoon, Chanyoung Park
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 changed | Where | Measured | Effect |
|---|---|---|---|
| apply DPL to Motifs | Motifs · VG150 PredCls | mR@50 / 100: 14.8 / 16.1 -> 33.7 / 37.4 | Improved |
| apply DPL to Motifs | Motifs · VG150 PredCls | R@50 / 100: 65.2 / 67.0 -> 54.4 / 56.3 | Hurt |
| apply DPL to Motifs on GQA | Motifs · GQA200 PredCls | mR@50 / 100: 14.9 / 15.6 -> 31.6 / 33.9 | Improved |
| use DPL instead of IETrans on VCTree | VCTree · VG150 PredCls | F@50 / 100: 38.6 / 41.9 -> 42.7 / 45.1 | Improved |
| add unbiased inference with sigma-normalized distance (ablation) | DPL · VG150 PredCls Motifs | F@50 / 100: 27.2 / 29.4 -> 41.6 / 44.9 | Improved |
| add orthogonal loss when using unbiased inference (ablation) | DPL · VG150 PredCls Motifs | mR@50 / 100: 24.7 / 26.9 -> 33.7 / 37.4 | Improved |
| add orthogonal loss alone without matching loss or unbiased inference (ablation) | DPL · VG150 PredCls Motifs | F@50 / 100: 27.4 / 29.3 -> 26.6 / 28.5 | No clear effect |
| increase number of samples N from 1 to 40 (ablation) | DPL · VG150 PredCls Motifs | R@50: 40.6 -> 56.8 | Improved |
| increase number of samples N from 1 to 40 (ablation) | DPL · VG150 PredCls Motifs | mR@50: 36.4 -> 33.0 | Hurt |
| set matching radius R away from 1.0 (ablation) | DPL · VG150 PredCls Motifs | F@50: 41.6 -> 31.5 (R=0.6) / 31.4 (R=1.4) | Hurt |
| assign more samples to head predicates than tail predicates (ablation) | DPL · VG150 PredCls Motifs | F@50 / 100: 37.7 / 40.5 -> 41.0 / 44.1 | Improved |
| reduce average N from 20 to 2 by setting N from predicate frequency (ablation) | DPL · VG150 PredCls Motifs | F@50 / 100: 41.3 / 44.6 -> 40.9 / 44.1 | No clear effect |
Closest papers by meaning
- RA-SGG: Retrieval-Augmented Scene Graph Generation Framework via Multi-Prototype Learning (AAAI 2025 · similarity 0.91)
- PPDL: Predicate Probability Distribution Based Loss for Unbiased Scene Graph Generation (CVPR 2022 · similarity 0.91)
- Panoptic Scene Graph Generation with Semantics-Prototype Learning (AAAI 2024 · similarity 0.91)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.89)
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 2024 · similarity 0.89)
- Multi-Prototype Space Learning for Commonsense-Based Scene Graph Generation (AAAI 2024 · similarity 0.89)
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.
