Research map / Graph learning / Knowledge and temporal graphs
Contextual Translation Embedding for Visual Relationship Detection and Scene Graph Generation
Zih-Siou Hung, Arun Mallya, Svetlana Lazebnik
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 norm regularization on embedding weights (C=1.0 vs C=0) (ablation) | UVTransE [V] · Stanford VRD (all test) | Relationship detection R@50: 4.67 -> 20.22 | Improved |
| increase C from 1.0 to 1.5 (ablation) | UVTransE [V] · Stanford VRD (all test) | Relationship detection R@50: 20.22 -> 19.99 | No clear effect |
| add spatial features to appearance features (ablation) | Appearance baseline · Stanford VRD | Predicate detection R@50 (all): 18.17 -> 38.89 | Improved |
| use subtraction instead of summation in translation embedding (ablation) | UVTransE [V] · Stanford VRD | Predicate detection R@50 (all): 49.01 -> 49.98 | Improved |
| add union box feature to translation formulation (UVTransE [V] vs VTransE [V]) (ablation) | VTransE · Stanford VRD (zero-shot) | Predicate detection R@50: 12.84 -> 22.92 | Improved |
| add union box feature to translation formulation (UVTransE [V] vs VTransE [V]) (ablation) | VTransE · Stanford VRD (all test) | Predicate detection R@50: 45.12 -> 49.98 | Improved |
| add Bi-GRU language module (ablation) | UVTransE · Stanford VRD (all test) | Predicate detection R@50: 49.98 -> 55.46 | Improved |
| add Bi-GRU language module (ablation) | UVTransE · Stanford VRD (zero-shot) | Predicate detection R@50: 22.92 -> 26.49 | Improved |
| add language module (ablation) | UVTransE · UnRel (with candidates, union) | mAP: 19.2 -> 18.0 | Hurt |
| add language module (ablation) | UVTransE (ResNet) · VG-IMP (SGGen) | Recall@50: 29.9 -> 30.1 | No clear effect |
| add language module (ablation) | UVTransE (VGG) · VG-IMP (SGGen) | Recall@50: 25.2 -> 25.3 | No clear effect |
| use ResNet-101 detector instead of VGG-16 (ablation) | UVTransE [V+L] · VG-IMP (SGGen) | Recall@50: 25.3 -> 30.1 | Improved |
Closest papers by meaning
- Visual Translation Embedding Network for Visual Relation Detection (CVPR 2017 · similarity 0.93)
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.90)
- LinkNet: Relational Embedding for Scene Graph (NeurIPS 2018 · similarity 0.90)
- Graphical Contrastive Losses for Scene Graph Parsing (CVPR 2019 · similarity 0.89)
- Relation-aware Hierarchical Prompt for Open-vocabulary Scene Graph Generation (AAAI 2025 · similarity 0.88)
- Leveraging Predicate and Triplet Learning for 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.
