Research map / Graph learning / Knowledge and temporal graphs
Fast Contextual Scene Graph Generation With Unbiased Context Augmentation
Tianlei Jin, Fangtai Guo, Qiwei Meng, Shiqiang Zhu, Xiangming Xi, Wen Wang, Zonghao Mu, Wei Song
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 |
|---|---|---|---|
| Use context-only predictions with context augmentation instead of a visual-feature SGG approach | C-SGG · VG PredCls and SGGen | Mean@K and F@K: 44.1/47.6 Mean@50/100 and 41.2/44.7 F@50/100; 20.8/23.8 Mean@50/100 and 19.0/21.9 F@50/100 | Improved |
| Add visual predicate analysis guided by contextual predictions | C-SGG · VG PredCls | R@50/100: 55.2/59.2 -> 58.2/62.4 | Improved |
| Add visual predicate analysis guided by contextual predictions | C-SGG · VG SGGen | mR@50/100: 14.8/17.1 -> 14.6/17.0 | No clear effect |
| Add visual predicate analysis guided by contextual predictions | C-SGG* · PSG SGGen | R@20/100: 24.0/29.0 -> 25.3/29.8 | Improved |
| Add visual predicate analysis guided by contextual predictions | C-SGG* · PSG SGGen | mR@20/100: 24.1/26.3 -> 23.0/25.8 | Hurt |
| Disable context augmentation (ε = 0) (ablation) | C-SGG · VG PredCls | mR@50/100: 32.9/36.0 -> 26.3/29.5 | Hurt |
| Set context augmentation factor to ε = 0.01 rather than ε = 0.05 (ablation) | C-SGG · VG PredCls | mR@50/100: 32.9/36.0 -> 27.8/30.2 | Hurt |
| Set context augmentation factor to ε = 0.1 rather than ε = 0.05 (ablation) | C-SGG · VG PredCls | Mean@50/100: 44.1/47.6 -> 43.6/45.1 | Hurt |
| Use ReLuL1 loss rather than BCE loss (ablation) | CV-SGG · VG PredCls | mR@50/100: 23.1/26.4 -> 32.6/36.2 | Improved |
| Set ReLuL1 boost factor to η = 0.2 rather than η = 0.1 (ablation) | CV-SGG · VG PredCls | mR@50/100: 32.6/36.2 -> 27.7/30.1 | Hurt |
| Use context experience in addition to visual predictions (α = 0.1) (ablation) | CV-SGG* · PSG SGGen | mR@20/100: 12.6/13.1 -> 14.8/16.3 | Improved |
| Set context-vision weight to α = 0.5 rather than α = 0.7 (ablation) | CV-SGG* · PSG SGGen | mR@20/100: 23.0/25.8 -> 22.3/24.8 | Hurt |
Closest papers by meaning
- Compositional Feature Augmentation for Unbiased Scene Graph Generation (ICCV 2023 · similarity 0.91)
- Expanding Scene Graph Boundaries: Fully Open-vocabulary Scene Graph Generation via Visual-Concept Alignment and Retention (ECCV 2024 · similarity 0.91)
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 2024 · 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 0.89)
- Adaptive Visual Scene Understanding: Incremental Scene Graph Generation (NeurIPS 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.
