Research map / Detection and segmentation / Open-set and few-shot detection
Compositional Learning for Human Object Interaction
Keizo Kato, Yin Li, Abhinav Gupta
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 external action links from knowledge base (ablation) | GCNCL-I · HICO unseen combos | mAP: 16.05 -> 16.82 | Improved |
| add new tokens (1 hop WordNet) on top of action links (ablation) | GCNCL-I+A · HICO unseen combos | mAP: 16.82 -> 16.99 | Improved |
| add external action links (ablation) | GCNCL · HICO unseen combos | mAP: 14.07 -> 15.65 | Improved |
| add new tokens on top of action links (ablation) | GCNCL+A · HICO unseen combos | mAP: 15.65 -> 16.35 | Improved |
| add verb-verb and noun-noun WordNet links to the graph (ablation) | GCNCL-I · HICO unseen combos | mAP: 16.05 -> 14.07 | Hurt |
| add verb-verb and noun-noun WordNet links to the graph (ablation) | GCNCL-I+NV+A · HICO novel concepts (unseen) | mAP: 6.92 -> 7.50 | Improved |
| add external action links (ablation) | GCNCL-I · HICO novel concepts (unseen) | mAP: 7.22 -> 6.73 | Hurt |
| add new tokens on top of action links (ablation) | GCNCL+A · HICO novel concepts (unseen) | mAP: 7.19 -> 7.50 | Improved |
| use GCNCL-I+NV+A instead of Triplet Siamese | Compositional model · HICO novel combos | mAP: 10.38 -> 11.51 | No clear effect |
| use GCNCL+NV+A instead of CC | Compositional model · HICO unseen combos | mAP: 15.98 -> 16.35 | Improved |
| use GCNCL+A instead of Triplet Siamese | Compositional model · Visual Genome HOI novel concepts (unseen) | mAP: 1.67 -> 2.44 | Improved |
| use GCNCL+A instead of Triplet Siamese | Compositional model · Charades novel concepts (unseen) | mAP: 7.82 -> 8.09 | No clear effect |
Closest papers by meaning
- Something-Else: Compositional Action Recognition With Spatial-Temporal Interaction Networks (CVPR 2020 · similarity 0.90)
- Discovering Human Interactions With Novel Objects via Zero-Shot Learning (CVPR 2020 · similarity 0.89)
- Why Can't I Open My Drawer? Mitigating Object-Driven Shortcuts in Zero-Shot Compositional Action Recognition (ECCV 2026 · similarity 0.89)
- Interaction Compass: Multi-Label Zero-Shot Learning of Human-Object Interactions via Spatial Relations (ICCV 2021 · similarity 0.89)
- C2C: Component-to-Composition Learning for Zero-Shot Compositional Action Recognition (ECCV 2024 · similarity 0.88)
- Visual Diversity and Region-aware Prompt Learning for Zero-shot HOI Detection (NeurIPS 2025 · 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.
