Research map / Detection and segmentation / Open-set and few-shot detection
Detecting Unseen Visual Relations Using Analogies
Julia Peyre, Ivan Laptev, Cordelia Schmid, Josef Sivic
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 visual phrase embedding to the compositional model (ablation) | s+o+p · HICO-DET (Default, full) | mAP: 18.7 -> 19.4 | Improved |
| add transfer by analogy (deep Gamma) to visual phrase embedding (ablation) | s+o+vp · HICO-DET 25 zero-shot triplets | mAP: 24.1 -> 28.6 | Improved |
| remove analogy transformation when aggregating source triplets (ablation) | s+o+vp+transfer · HICO-DET 25 zero-shot triplets | mAP: 9.6 -> 28.6 | Improved |
| use 2-layer MLP instead of linear Gamma (ablation) | s+o+vp+transfer · HICO-DET 25 zero-shot triplets | mAP: 27.6 -> 28.6 | Improved |
| use learned Gamma instead of zero transformation (ablation) | s+o+vp+transfer · HICO-DET 25 zero-shot triplets | mAP: 24.8 -> 28.6 | Improved |
| add transfer by analogy (ablation) | p+vp · UnRel (IoU=0.3, union) | mAP: 16.4 -> 17.5 | Improved |
| add visual phrase embedding without transfer (ablation) | p · UnRel (IoU=0.3, union) | mAP: 16.8 -> 16.4 | No clear effect |
| add transfer by analogy (ablation) | s+o+vp · COCO-a unseen triplets (all) | mAP: 6.0 -> 6.9 | Improved |
| add transfer by analogy (ablation) | s+o+p+vp · COCO-a unseen triplets (all) | mAP: 5.1 -> 5.2 | No clear effect |
| add bigram branches (ablation) | s+o+p+vp · HICO-DET (Default, full) | mAP: 19.4 -> 20.0 | Improved |
| map triplets to embedding space instead of learning s, o, p classifiers (ablation) | s+o+p · HICO-DET (Default, full) | mAP: 18.0 -> 18.4 | Improved |
| use pre-trained word embeddings instead of random words (ablation) | s+o+p · HICO-DET (Default, full) | mAP: 18.4 -> 18.7 | Improved |
Closest papers by meaning
- Weakly-Supervised Learning of Visual Relations (ICCV 2017 · similarity 0.92)
- Explanation-based Weakly-supervised Learning of Visual Relations with Graph Networks (ECCV 2020 · similarity 0.89)
- Visual Relationship Detection with Language Priors (ECCV 2016 · similarity 0.89)
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.89)
- Visual Relationship Detection With Internal and External Linguistic Knowledge Distillation (ICCV 2017 · similarity 0.89)
- Leveraging Predicate and Triplet Learning for Scene Graph Generation (CVPR 2024 · similarity 0.87)
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.
