Research map / Graph learning / Knowledge and temporal graphs
SPAN: Learning Similarity Between Scene Graphs and Images With Transformers
Yuren Cong, Wentong Liao, Bodo Rosenhahn, Michael Ying Yang
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 |
|---|---|---|---|
| replace GCN with graph Transformer encoder (ablation) | SPAN · Visual Genome (LF graph) | R-Precision K=100: 46.2 -> 58.0 | Improved |
| replace GCN with graph Transformer encoder (ablation) | SPAN · Visual Genome (LB graph) | R-Precision K=100: 63.2 -> 71.0 | Improved |
| add structural encoding to graph serialization (ablation) | SPAN · Visual Genome (LF graph) | R-Precision K=100: 47.8 -> 58.0 | Improved |
| add structural encoding to graph serialization (ablation) | SPAN · Visual Genome (LB graph) | R-Precision K=100: 65.1 -> 71.0 | Improved |
| add node shuffle during training (ablation) | SPAN · Visual Genome (LF graph) | R-Precision K=100: 52.8 -> 58.0 | Improved |
| add node shuffle during training (ablation) | SPAN · Visual Genome (LB graph) | R-Precision K=100: 68.3 -> 71.0 | Improved |
| add edge embeddings to node embeddings (ablation) | SPAN · Visual Genome (LF graph) | R-Precision K=100: 46.4 -> 58.0 | Improved |
| use only edge embeddings instead of full graph (ablation) | SPAN-Edge · Visual Genome (LF graph) | R-Precision K=100: 58.0 -> 1.8 | Hurt |
| add bounding boxes to scene graph query (LB vs LF) | SPAN · Visual Genome | R-Precision K=100: 58.0 -> 71.0 | Improved |
| use SPAN scene graph query instead of CLIP text of triplets | SPAN · Visual Genome (LF graph) | R-Precision K=100: 47.2 -> 58.0 | Improved |
| increase Transformer layers from 6/6 to 8/8 (ablation) | SPAN · Visual Genome (LF graph) | R-Precision K=100: 58.0 -> 56.3 | Hurt |
| reduce Transformer layers from 6/6 to 4/4 (ablation) | SPAN · Visual Genome (LF graph) | R-Precision K=100: 58.0 -> 54.9 | Hurt |
Closest papers by meaning
- Scene Graph Embeddings Using Relative Similarity Supervision (AAAI 2021 · similarity 0.90)
- LinkNet: Relational Embedding for Scene Graph (NeurIPS 2018 · similarity 0.89)
- Scene Graph Guided Generation: Enable Accurate Relations Generation in Text-to-Image Models via Textural Rectification (ICCV 2025 · similarity 0.89)
- A Comprehensive Survey of Scene Graphs: Generation and Application (TPAMI 2021 · similarity 0.89)
- SCENIR: Visual Semantic Clarity through Unsupervised Scene Graph Retrieval (ICML 2025 · similarity 0.88)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 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.
