Research map / Graph learning / Knowledge and temporal graphs
Expanding Scene Graph Boundaries: Fully Open-vocabulary Scene Graph Generation via Visual-Concept Alignment and Retention
Zuyao Chen, Jinlin Wu, Zhen Lei, Zhaoxiang Zhang, Chang Wen Chen
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 knowledge-distillation visual-concept retention (ablation) | OvSGTR (Swin-T) · VG150 OvR-SGG, Novel relation | R@50: 0.34 -> 13.45 | Improved |
| add knowledge-distillation visual-concept retention (ablation) | OvSGTR (Swin-B) · VG150 OvD+R-SGG, Novel relation | R@50: 1.78 -> 14.56 | Improved |
| add knowledge-distillation visual-concept retention (ablation) | OvSGTR (Swin-T) · VG150 OvD+R-SGG, Joint Base+Novel | R@50: 7.88 -> 13.53 | Improved |
| set distillation weight lambda to 0.1 versus 0 (ablation) | OvSGTR on OvR-SGG · VG150 val OvR-SGG, Novel | R@50: 0.32 -> 11.54 | Improved |
| increase lambda from 0.1 to 0.5 (ablation) | OvSGTR on OvR-SGG · VG150 val OvR-SGG, Base+Novel | R@50: 16.00 -> 13.34 | Hurt |
| use a single relation query rather than none or multiple (ablation) | OvSGTR · VG150 val Closed-set SGG | Recall | Improved |
| replace Swin-T with Swin-B backbone | OvSGTR · VG150 OvD-SGG, Novel object SGDET | R@50: 12.06 -> 15.58 | Improved |
| enlarge caption pre-training data to COCO+Flickr30k+SBU | OvSGTR (Swin-T) · VG150 test language-supervised SGG | R@50: 8.92 -> 9.43 | Improved |
| use DETR-style one-stage model with two-layer MLP relation head instead of message passing | OvSGTR (Swin-T) vs VS3 (Swin-L) · VG150 Closed-set SGG | R@50 and latency: 0.13 s vs. 0.24 s; 41M vs. 124M trainable params | Improved |
Closest papers by meaning
- From Pixels to Graphs: Open-Vocabulary Scene Graph Generation with Vision-Language Models (CVPR 2024 · similarity 0.94)
- Relation-aware Hierarchical Prompt for Open-vocabulary Scene Graph Generation (AAAI 2025 · similarity 0.92)
- Scene Graph Generation with Role-Playing Large Language Models (NeurIPS 2024 · similarity 0.92)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 2024 · similarity 0.90)
- Adaptive Visual Scene Understanding: Incremental Scene Graph Generation (NeurIPS 2024 · similarity 0.90)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (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.
