Research map / Detection and segmentation / Open-set and few-shot detection
Visual Relationship Detection with Language Priors
Cewu Lu, Ranjay Krishna, Michael Bernstein, 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 K() embedding term that pulls similar relationships together, on top of V+L (ablation) | Ours - V + L · Visual Relationship Detection dataset, predicate detection | Recall@100: 35.20 -> 47.87 | Improved |
| add K() embedding term on top of V+L (ablation) | Ours - V + L · Visual Relationship Detection dataset, phrase detection | Recall@100: 8.59 -> 17.03 | Improved |
| add K() embedding term on top of V+L (ablation) | Ours - V + L · Visual Relationship Detection dataset, relationship detection | Recall@100: 9.18 -> 14.70 | Improved |
| use K() instead of a plain L2 regularizer on W (ablation) | Ours - V + L + Reg. · Visual Relationship Detection dataset, relationship detection | Recall@100: 9.63 -> 14.70 | Improved |
| add language likelihood L to the visual module (ablation) | Ours - V only · Visual Relationship Detection dataset, predicate detection | Recall@100: 7.11 -> 35.20 | Improved |
| add visual module V to the language-only model (ablation) | Ours - L only · Visual Relationship Detection dataset, relationship detection | Recall@100: 0.08 -> 9.18 | Improved |
| use learned projection function f() instead of naive training-frequency function (ablation) | Ours - V + naive FC · Visual Relationship Detection dataset, predicate detection | Recall@100: 28.87 -> 35.20 | Improved |
| add K() embedding term for unseen relationships (ablation) | Ours - V + L · Visual Relationship Detection dataset (zero-shot), relationship detection | Recall@100: 2.66 -> 3.52 | Improved |
| add K() embedding term for unseen relationships (ablation) | Ours - V + L · Visual Relationship Detection dataset (zero-shot), predicate detection | Recall@100: 6.11 -> 8.45 | Improved |
| add language likelihood L to visual module for unseen relationships (ablation) | Ours - V only · Visual Relationship Detection dataset (zero-shot), phrase detection | Recall@100: 1.12 -> 2.56 | Improved |
| add K() embedding term on Visual Phrases dataset (ablation) | Ours V + L only · Visual Phrases dataset (12 phrases), phrase detection | mAP: 57.3 -> 59.2 | Improved |
| add language likelihood L on Visual Phrases dataset (ablation) | Ours V only · Visual Phrases dataset (12 phrases), phrase detection | mAP: 53.4 -> 57.3 | Improved |
Closest papers by meaning
- Visual Relationship Detection With Internal and External Linguistic Knowledge Distillation (ICCV 2017 · similarity 0.91)
- Explanation-based Weakly-supervised Learning of Visual Relations with Graph Networks (ECCV 2020 · similarity 0.91)
- Large-Scale Visual Relationship Understanding (AAAI 2019 · similarity 0.91)
- Tensorize, Factorize and Regularize: Robust Visual Relationship Learning (CVPR 2018 · similarity 0.91)
- Scene-Graph ViT: End-to-End Open-Vocabulary Visual Relationship Detection (ECCV 2024 · similarity 0.91)
- Multi-Modal Prompting for Open-Vocabulary Video Visual Relationship Detection (AAAI 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.
