Research map / Graph learning / Knowledge and temporal graphs
Improving Scene Graph Classification by Exploiting Knowledge from Texts
Read the paperCheck 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 |
|---|---|---|---|
| Fine-tune the relational reasoning module with text-extracted graphs (ablation) | SCH scene graph classifier · Visual Genome test, SGCls, 1% | R@100: 15.75±0.64 -> 19.42 ± 0.47 | Helped |
| Fine-tune the relational reasoning module with text-extracted graphs (ablation) | SCH scene graph classifier · Visual Genome test, SGCls, 10% | R@100: 33.94±0.94 -> 35.03 ± 0.49 | Helped |
| Fine-tune the relational reasoning module with text-extracted graphs (ablation) | SCH predicate classifier · Visual Genome test, PredCls, 1% | R@100: 63.64±0.75 -> 79.17 ± 0.16 | Helped |
| Fine-tune the relational reasoning module with text-extracted graphs (ablation) | SCH predicate classifier · Visual Genome test, PredCls, 10% | R@100: 76.85±0.38 -> 80.54 ± 0.10 | Helped |
| Fine-tune the relational reasoning module with text-extracted graphs (ablation) | SCH object classifier · Visual Genome test, object classification, 1% | Top-1 accuracy: 40.75±0.48 -> 42.70 ± 0.49 | Helped |
| Fine-tune the relational reasoning module with text-extracted graphs (ablation) | SCH object classifier · Visual Genome test, object classification, 10% | Top-1 accuracy: 56.97±0.76 -> 58.31 ± 0.45 | Helped |
| Use T5-predicted graphs rather than ground-truth graphs for fine-tuning | Visual Genome scene graph classifier · Visual Genome test, SGCls, 1% | R@100: 19.42 ± 0.47 vs 19.59±0.49 | No clear effect |
| Use T5-predicted graphs rather than ground-truth graphs for fine-tuning | Visual Genome predicate classifier · Visual Genome test, PredCls, 1% | R@100: 79.17 ± 0.16 vs 84.30±0.16 | Hurt |
| Use T5-predicted graphs rather than ground-truth graphs for fine-tuning | Visual Genome object classifier · Visual Genome test, object classification, 1% | Top-1 accuracy: 42.70 ± 0.49 vs 42.09±0.65 | Helped |
| Use T5 instead of CopyNet for text-to-graph extraction | Visual Genome text-to-graph model · Visual Genome text-to-graph, 1% | F1: 28.16±0.21 -> 32.17±0.13 | Helped |
Closest papers by meaning
- Classification by Attention: Scene Graph Classification with Prior Knowledge (AAAI 2021 · similarity 0.92)
- Scene Graph Generation with Role-Playing Large Language Models (NeurIPS 2024 · similarity 0.89)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.89)
- Object-Centric Representation Learning for Enhanced 3D Semantic Scene Graph Prediction (NeurIPS 2025 · similarity 0.88)
- SCENIR: Visual Semantic Clarity through Unsupervised Scene Graph Retrieval (ICML 2025 · similarity 0.87)
- Structure-CLIP: Towards Scene Graph Knowledge to Enhance Multi-Modal Structured Representations (AAAI 2024 · similarity 0.87)
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.