Research map / Graph learning / Knowledge and temporal graphs
Learning Visual Commonsense for Robust Scene Graph Generation
Alireza Zareian, Zhecan Wang, Haoxuan You, Shih-Fu Chang
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 |
|---|---|---|---|
| add local attention heads to a global-only transformer (GLAT) (ablation) | scene graph transformer · VG masked reconstruction | overall node accuracy (%): 22.9 -> 34.4 | Improved |
| add global heads to a local-only graph attention net (GLAT) (ablation) | scene graph transformer · VG masked reconstruction | overall node accuracy (%): 22.3 -> 34.4 | Improved |
| learned attention instead of fixed equal attention on neighbors (ablation) | graph network · VG masked reconstruction | overall node accuracy (%): 19.7 -> 22.3 | Improved |
| GLAT instead of triplet frequency prior for masked predicates | commonsense model · VG masked reconstruction | predicate accuracy (%): 44.4 -> 60.7 | Improved |
| stack GLAT on IMP | IMP · VG SGG PredCls | mR@50: 9.8 -> 11.1 | Improved |
| add fusion module to IMP + GLAT (ablation) | IMP + GLAT · VG SGG PredCls | mR@50: 11.1 -> 12.1 | Improved |
| stack GLAT on SNM | SNM · VG SGG PredCls | mR@50: 13.3 -> 13.6 | No clear effect |
| add fusion module to SNM + GLAT (ablation) | SNM + GLAT · VG SGG PredCls | mR@50: 13.6 -> 14.1 | Improved |
| stack GLAT on KERN | KERN · VG SGG PredCls | mR@50: 17.7 -> 17.6 | No clear effect |
| add fusion module to KERN + GLAT (ablation) | KERN + GLAT · VG SGG SGCls | mR@50: 9.3 -> 9.9 | Improved |
| add GLAT and fusion to IMP | IMP · VG SGG SGCls | mR@100: 6.0 -> 7.0 | Improved |
| add GLAT and fusion to IMP | IMP · VG SGCls R@100 by triplet frequency (average) | R@100 (%): 36.1 -> 37.4 | Improved |
Closest papers by meaning
- Bridging Knowledge Graphs to Generate Scene Graphs (ECCV 2020 · similarity 0.93)
- SGEITL: Scene Graph Enhanced Image-Text Learning for Visual Commonsense Reasoning (AAAI 2022 · similarity 0.91)
- APT: Towards Universal Scene Graph Generation via Plug-in Adaptive Prompt Tuning (ICLR 2026 · similarity 0.90)
- A Comprehensive Survey of Scene Graphs: Generation and Application (TPAMI 2021 · similarity 0.90)
- Scene Graph Thinking: Reinforcing Structured Visual Reasoning for Multimodal Large Language Models (ICML 2026 · similarity 0.89)
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 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.
