Research map / Detection and segmentation / Weakly supervised detection and segmentation
Weakly-Supervised Learning of Visual Relations
Julia Peyre, Josef Sivic, Ivan Laptev, Cordelia Schmid
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 spatial features to appearance features (ablation) | Ours (full sup.) · VRD predicate detection (All) | recall@50: 46.3 -> 50.4 | Improved |
| add appearance features to spatial features (ablation) | Ours (full sup.) · VRD predicate detection (All) | recall@50: 42.2 -> 50.4 | Improved |
| add spatial features to appearance features (ablation) | Ours (full sup.) · VRD predicate detection (Unseen) | recall@50: 16.1 -> 23.6 | Improved |
| add language scores of Lu et al. | Ours [S+A] (full sup.) · VRD predicate detection (All vs Unseen) | recall@50: 50.4 -> 52.6 (All); 23.6 -> 21.6 (Unseen) | Mixed |
| switch from full to weak image-level supervision | Ours [S+A] · VRD predicate detection (All) | recall@50: 50.4 -> 46.8 | Hurt |
| switch from full to weak image-level supervision | Ours [S+A] · VRD predicate detection (Unseen) | recall@50: 23.6 -> 19.0 | Hurt |
| infer which object pairs are in relation (discriminative clustering) instead of picking a random pair (ablation) | Ours [S+A] weak sup. · VRD relationship detection (Unseen) | recall@50: 5.6 -> 6.7 | Improved |
| infer which object pairs are in relation instead of picking a random pair, with ~5x more candidate pairs (ablation) | Ours [S+A] weak sup. · VRD harder candidate setup, phrase detection | recall@50: 15.3 -> 17.9 | Improved |
| infer which object pairs are in relation instead of picking a random pair, with ~5x more candidate pairs (ablation) | Ours [S+A] weak sup. · VRD harder candidate setup, relationship detection | recall@50: 10.1 -> 12.0 | Improved |
| infer object pairs instead of picking a random pair (ablation) | Ours [S+A] weak sup. · UnRel retrieval (with GT boxes) | mAP: 55.0 -> 58.5 | Improved |
| use spatial plus appearance visual representation instead of union-box appearance classifier | Ours [S+A] full sup. · UnRel retrieval (with GT boxes) | mAP: 50.6 -> 62.6 | Improved |
| use spatial plus appearance visual representation instead of union-box appearance classifier | Ours [S+A] full sup. · Visual Genome relationship recognition | top-1 accuracy: 8.7 -> 46.5 | Improved |
Closest papers by meaning
- Explanation-based Weakly-supervised Learning of Visual Relations with Graph Networks (ECCV 2020 · similarity 0.93)
- Detecting Unseen Visual Relations Using Analogies (ICCV 2019 · similarity 0.92)
- Localize, Assemble, and Predicate: Contextual Object Proposal Embedding for Visual Relation Detection (AAAI 2020 · similarity 0.91)
- Visual Relationship Detection with Language Priors (ECCV 2016 · similarity 0.90)
- Visual Relationship Detection With Internal and External Linguistic Knowledge Distillation (ICCV 2017 · similarity 0.90)
- 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.
