Research map / Graph learning / Knowledge and temporal graphs
Devil's on the Edges: Selective Quad Attention for Scene Graph Generation
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 |
|---|---|---|---|
| Use SQUAT instead of BGNN | Visual Genome scene graph generation · Visual Genome SGDet | mR@100: 12.6 -> 16.5 | Helped |
| Add edge selection to quad attention (ablation) | SQUAT · Visual Genome SGDet | mR@100: 15.00 -> 16.47 | Helped |
| Use query edge selection (ablation) | SQUAT · Visual Genome SGDet | mR@100: 15.00 -> 15.66 | Helped |
| Use distinct Q, E2E, and N2E selection modules instead of shared parameters (ablation) | SQUAT · Visual Genome SGDet | mR@100: 14.85 -> 16.47 | Helped |
| Add edge-to-node and edge-to-edge attention (ablation) | SQUAT · Visual Genome SGDet | mR@100: 11.57 -> 16.47 | Helped |
| Add node-to-node and node-to-edge attention (ablation) | SQUAT · Visual Genome SGDet | mR@100: 15.28 -> 16.47 | Helped |
| Add edge-to-edge attention (ablation) | SQUAT · Visual Genome SGDet | mR@100: 15.03 -> 16.47 | Helped |
| Use message passing over a fully connected graph (ablation) | IMP · Visual Genome SGDet | mR@100: 6.53 -> 5.42 | Hurt |
| Use message passing over a fully connected graph (ablation) | BGNN · Visual Genome SGDet | mR@100: 13.56 -> 12.46 | Hurt |
| Apply edge selection to BGNN (ablation) | BGNN · Visual Genome SGDet | mR@100: 13.56 -> 14.20 | Helped |
| Use ground-truth edges for message passing (ablation) | SQUAT · Visual Genome SGDet | mR@100: 13.99 -> 19.51 | Helped |
| Use ground-truth edges for message passing (ablation) | BGNN · Visual Genome SGDet | mR@100: 13.56 -> 17.09 | Helped |
Closest papers by meaning
- Scene Graph Generation by Iterative Message Passing (CVPR 2017 · similarity 0.89)
- LinkNet: Relational Embedding for Scene Graph (NeurIPS 2018 · similarity 0.88)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.87)
- HyperGLM: HyperGraph for Video Scene Graph Generation and Anticipation (CVPR 2025 · similarity 0.87)
- Relation-aware Hierarchical Prompt for Open-vocabulary Scene Graph Generation (AAAI 2025 · similarity 0.86)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.86)
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.