Research map / Graph learning / Knowledge and temporal graphs
A Simple Baseline for Weakly-Supervised Scene Graph Generation
Jing Shi, Yiwu Zhong, Ning Xu, Yin Li, Chenliang Xu
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 |
|---|---|---|---|
| train SGG with WSGM pseudo labels instead of VSPNet (EAMP) | SGG model · VG split SGGen | Recall@100: 4.17 -> 6.02 | Improved |
| use first-order graph matching WSGM instead of VSPNet second-order | graph matching · VG train set SGGen | Rinst: 2.75 -> 9.07 | Improved |
| use first-order graph matching WSGM instead of VSPNet | graph matching · VG train set SGCls | Rinst: 59.89 -> 67.61 | Improved |
| use triplet loss instead of logistic loss (ablation) | WSGM · VG SGGen | Recall@100: 2.01 -> 4.56 | Improved |
| use triplet loss instead of NCE loss (ablation) | WSGM · VG SGGen | Recall@100: 2.17 -> 4.56 | Improved |
| anchor on label node (L_vis) instead of visual node (ablation) | WSGM · VG SGGen | Recall@100: 5.43 -> 6.02 | Improved |
| combine both contrast directions (ablation) | WSGM · VG SGGen | Recall@100: 6.02 -> 5.39 | Hurt |
| use one-to-one graph matching instead of independent grounding (ablation) | WSGM · VG SGGen | Recall@100: 4.48 -> 6.02 | Improved |
| keep whole graph instead of sub-graph with 10% nodes (ablation) | WSGM · VG SGGen | Recall@100: 4.79 -> 6.02 | Improved |
| keep whole graph instead of sub-graph with 50% nodes (ablation) | WSGM · VG SGGen | Recall@100: 5.81 -> 6.02 | Improved |
| add CrossMatch negatives from other images (ablation) | WSGM · VG SGGen | Recall@100: 4.56 -> 6.02 | Improved |
| add HardNeg mining (ablation) | WSGM · VG SGGen | Recall@100: 5.31 -> 6.02 | Improved |
Closest papers by meaning
- Visual Distant Supervision for Scene Graph Generation (ICCV 2021 · similarity 0.91)
- Revisiting Weakly-Supervised Video Scene Graph Generation via Pair Affinity Learning (ECCV 2026 · similarity 0.91)
- Scene Graph Prediction With Limited Labels (ICCV 2019 · similarity 0.90)
- Weakly Supervised Visual Semantic Parsing (CVPR 2020 · similarity 0.90)
- Iterative Scene Graph Generation (NeurIPS 2022 · similarity 0.90)
- DSGG: Dense Relation Transformer for an End-to-end Scene Graph Generation (CVPR 2024 · similarity 0.88)
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.
