Research map / Graph learning / Knowledge and temporal graphs
Scene Graph Prediction With Limited Labels
Vincent S. Chen, Paroma Varma, Ranjay Krishna, Michael Bernstein, Christopher Re, Li Fei-Fei
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 |
|---|---|---|---|
| add spatial features to categorical features (ablation) | label generation · Visual Genome (20 predicates) | PREDCLS recall@100: 45.50 -> 47.53 | Improved |
| add categorical features to spatial features (ablation) | label generation · Visual Genome (20 predicates) | PREDCLS recall@100: 27.26 -> 47.53 | Improved |
| add deep ResNet50 image features to categorical + spatial (ablation) | label generation · Visual Genome (20 predicates) | PREDCLS recall@100: 47.53 -> 40.02 | Hurt |
| add word vectors for categorical features (ablation) | label generation · Visual Genome (20 predicates) | PREDCLS recall@100: 47.32 -> 47.53 | No clear effect |
| use factor-graph generative model instead of majority vote (ablation) | label aggregation · Visual Genome (20 predicates) | SGDET recall@100: 18.57 -> 19.28 | Improved |
| use generative model instead of majority vote (ablation) | label aggregation · VRD label quality (n=10) | F1: 56.11 -> 57.66 | Improved |
| use generated labels versus labeled examples only (ablation) | scene graph model · Visual Genome (20 predicates) | PREDCLS recall@100: 6.61 -> 47.53 | Improved |
| semi-supervised generated labels instead of transfer learning | scene graph model · Visual Genome (20 predicates) | PREDCLS recall@100: 42.37 -> 47.53 | Improved |
| generated labels instead of single decision tree labels | scene graph model · Visual Genome (20 predicates) | PREDCLS recall@100: 33.35 -> 47.53 | Improved |
| generated labels instead of label propagation | scene graph model · Visual Genome (20 predicates) | PREDCLS recall@100: 25.41 -> 47.53 | Improved |
| reduce labeled examples per predicate toward n=10 | Ours vs Transfer Learning · Visual Genome (20 predicates) | PREDCLS recall@100 | Improved |
| use generated labels with 50 predicates | scene graph model · Visual Genome (top 50 predicates) | PREDCLS recall@100: 13.07 -> 28.53 | Improved |
Closest papers by meaning
- Visual Distant Supervision for Scene Graph Generation (ICCV 2021 · similarity 0.93)
- Scene Graph Generation by Iterative Message Passing (CVPR 2017 · similarity 0.91)
- Not All Relations Are Equal: Mining Informative Labels for Scene Graph Generation (CVPR 2022 · similarity 0.91)
- Learning To Generate Scene Graph From Natural Language Supervision (ICCV 2021 · similarity 0.91)
- A Simple Baseline for Weakly-Supervised Scene Graph Generation (ICCV 2021 · similarity 0.90)
- DSGG: Dense Relation Transformer for an End-to-end 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.
