Research map / Vision-language and multimodal / Human and embodied modeling
Detecting Visual Relationships With Deep Relational Networks
Check 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 |
|---|---|---|---|
| Replace VR with DR-Net | Visual relationship detection · VRD predicate recognition | Recall@50: 47.87 -> 80.78 | Helped |
| Replace VR with DR-Net | Visual relationship detection · sVG predicate recognition | Recall@50: 53.49 -> 88.26 | Helped |
| Add pair filter to DR-Net (ablation) | DR-Net · VRD union box detection | Recall@50: 19.02 -> 19.93 | Helped |
| Add spatial module to appearance-only A1 (ablation) | A1 (VGG16 appearance) · VRD predicate recognition | Recall@50: 63.39 -> 71.81 | Helped |
| Add DR-Net relational modeling to A1 plus spatial module (ablation) | A1S · VRD predicate recognition | Recall@50: 71.81 -> 80.66 | Helped |
| Use CRF rather than DR-Net for statistical dependencies (ablation) | A1 plus spatial module · VRD predicate recognition | Recall@50: 72.77 vs. 80.66 | Hurt |
| Replace VGG16 with ResNet101 (ablation) | A1SD · VRD predicate recognition | Recall@50: 80.66 -> 80.78 | Helped |
| Add spatial module without appearance features (ablation) | S-only model · VRD predicate recognition | Recall@50: 64.72 vs. 63.39; 75.18 vs. 72.13 | Mixed |
| Add DR-Net to A1 plus spatial module (ablation) | A1S · sVG predicate recognition | Recall@50: 79.10 -> 88.00 | Helped |
| Use unshared relational weights across inference units (ablation) | DR-Net · sVG | recall: the network without weight sharing performs relatively better | Helped |
| Increase inference units (ablation) | DR-Net · sVG | recall: 56% to 73% | Helped |
| Increase spatial mask size from 8 to 32 (ablation) | Dual spatial masks · VRD | top-1 recall: 0.47 -> 0.50 | Helped |
Closest papers by meaning
- Visual Relationship Detection With Deep Structural Ranking (AAAI 2018 · similarity 0.91)
- Large-Scale Visual Relationship Understanding (AAAI 2019 · similarity 0.91)
- MR-NET: Exploiting Mutual Relation for Visual Relationship Detection (AAAI 2019 · similarity 0.90)
- Deep Structured Learning for Visual Relationship Detection (AAAI 2018 · similarity 0.90)
- Tensorize, Factorize and Regularize: Robust Visual Relationship Learning (CVPR 2018 · similarity 0.90)
- Relationship Proposal Networks (CVPR 2017 · 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.