Research map / Detection and segmentation / Open-set and few-shot detection
Visual Translation Embedding Network for Visual Relation Detection
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 |
|---|---|---|---|
| Use translation embedding for predicate prediction instead of JointBox classification | VTransE predicate prediction · VRD | R@50: 25.78 -> 44.76 | Helped |
| Use translation embedding for predicate prediction instead of JointBox classification | VTransE predicate prediction · VG | R@50: 46.59 -> 62.63 | Helped |
| Train object detection jointly end-to-end with relation prediction (ablation) | VTransE · VRD | object detection mAP%: 13.32 -> 13.98 | Helped |
| Train object detection jointly end-to-end with relation prediction (ablation) | VTransE · VG | object detection mAP%: 6.21 -> 6.58 | Helped |
| Use all features—classeme, location, and visual—with learned scaling rather than a single feature type (ablation) | VTransE · VRD and VG | relation detection R@100 | Helped |
| Use location features rather than other individual feature types for spatial relations (ablation) | VTransE · VRD and VG | relation detection R@100 | Helped |
| Use classeme features rather than visual features alone (ablation) | VTransE · VRD and VG | relation detection R@100 | Helped |
| Train jointly end-to-end rather than in two stages (ablation) | VTransE · VRD and VG | relation detection R@50: 13.30 -> 14.07; 4.97 -> 5.52 | Helped |
| Train jointly end-to-end rather than in two stages (ablation) | VTransE · VRD and VG | relation detection R@100: 14.64 -> 15.20; 5.48 -> 6.04 | Helped |
| Use a purely visual model instead of a model with language priors | VTransE versus Lu's-VLK · VRD zero-shot | relation detection R@100: 3.52 -> 2.14 | Hurt |
| Use a purely visual model instead of a model with language priors | VTransE versus Lu's-VLK · VRD zero-shot | relation retrieval Rr@5: 1.24 -> 1.42 | Helped |
Closest papers by meaning
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.88)
- Zero-shot Visual Relation Detection via Composite Visual Cues from Large Language Models (NeurIPS 2023 · similarity 0.87)
- Detecting Visual Relationships With Deep Relational Networks (CVPR 2017 · similarity 0.87)
- RelationVGGT : Visual Geometry Transformers for 3D Spatial Relation Segmentation (NeurIPS 2026 · similarity 0.85)
- Relation Rectification in Diffusion Model (CVPR 2024 · similarity 0.84)
- EGTR: Extracting Graph from Transformer for Scene Graph Generation (CVPR 2024 · similarity 0.84)
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.