Research map / Graph learning / Knowledge and temporal graphs
Unbiased Heterogeneous Scene Graph Generation with Relation-Aware Message Passing Neural Network
Kanghoon Yoon, Kibum Kim, Jinyoung Moon, Chanyoung Park
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 heterogeneous relation-aware message passing instead of homogeneous BGNN message passing | SGG models · Visual Genome PredCls | mR@50/100: 30.4/32.9 -> 31.6/33.5 | Improved |
| use heterogeneous relation-aware message passing instead of homogeneous BGNN message passing | SGG models · Visual Genome SGCls | mR@50/100: 14.3/16.5 -> 17.2/18.7 | Improved |
| use heterogeneous relation-aware message passing instead of homogeneous BGNN message passing | SGG models · Visual Genome SGGen | mR@50/100: 10.7/12.6 -> 12.2/14.4 | Improved |
| use heterogeneous relation-aware message passing instead of BGNN | SGG models · Open Images V6 SGGen | mR@50: 40.5 -> 42.7 | Improved |
| use heterogeneous relation-aware message passing instead of BGNN | SGG models · Open Images V6 SGGen | R@50: 75.0 -> 76.8 | Improved |
| add relation-type-specific weights to the edge-wise update (ablation) | HetSGG · Visual Genome SGCls | mR@100: 15.9 -> 16.2 | Improved |
| add relation-type-specific weights to the node-wise update (ablation) | HetSGG · Visual Genome SGCls | mR@100: 15.9 -> 17.7 | Improved |
| add relation-type-specific weights to both edge-wise and node-wise updates (ablation) | HetSGG · Visual Genome SGCls | mR@100: 15.9 -> 18.7 | Improved |
| use 8 basis matrices rather than 4 (ablation) | HetSGG · Visual Genome SGCls | mR@100: 17.2 -> 18.7 | Improved |
| use 8 basis matrices rather than 12 (ablation) | HetSGG · Visual Genome SGCls | mR@100: 17.6 -> 18.7 | Improved |
| use 8 basis matrices rather than 16 (ablation) | HetSGG · Visual Genome SGCls | mR@100: 18.2 -> 18.7 | Improved |
| use ground-truth object types rather than inferred types (ablation) | HetSGG with three object types · Visual Genome SGCls | mR@50/100: 17.2 / 18.7 -> 17.4 / 19.1 | Improved |
Closest papers by meaning
- HL-Net: Heterophily Learning Network for Scene Graph Generation (CVPR 2022 · similarity 0.93)
- Scene Graph Generation by Iterative Message Passing (CVPR 2017 · similarity 0.91)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.91)
- Scene Graph-Grounded Image Generation (AAAI 2025 · similarity 0.90)
- Scene Graph Generation Strategy with Co-occurrence Knowledge and Learnable Term Frequency (ICML 2024 · similarity 0.90)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 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.
