Research map / Detection and segmentation / Open-set and few-shot detection
Exploring Predicate Visual Context in Detecting of Human-Object Interactions
Frederic Z Zhang, Yuhui Yuan, Dylan Campbell, Zhuoyao Zhong, Stephen Gould
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 cross-attention with frozen encoder features as keys/values (ablation) | UPT-like two-stage detector · HICO-DET (Default, Full) | mAP: 31.47 -> 31.51 | No clear effect |
| use backbone C5 features instead of encoder features as cross-attention keys/values (ablation) | PViC · HICO-DET (Default, Full) | mAP: 31.51 -> 32.89 | Improved |
| use higher-resolution or lower-level features (C3, C4, FPN P3) as keys/values (ablation) | PViC · HICO-DET (Default, Full) | mAP: 32.89 -> 29.59 (C3 with head) | Hurt |
| add encoder features alongside C5 keys/values (ablation) | PViC · HICO-DET (Default, Full) | mAP: 32.89 -> 32.69 | No clear effect |
| add a lightweight window-attention feature head on C5 (ablation) | PViC · HICO-DET (Default, Full) | mAP: 32.89 -> 33.54 | Improved |
| stack more attention layers in feature head (2x vs 1x) (ablation) | PViC · HICO-DET (Default, Full) | mAP: 33.54 -> 33.57 | No clear effect |
| replace UPT modified self-attention with vanilla self-attention plus box positional embeddings (ablation) | PViC · HICO-DET (Default, Full) | mAP: 33.54 -> 33.59 | No clear effect |
| add spatial features in query fusion (with no self-attn) (ablation) | PViC explicit queries · HICO-DET (Default, Full) | mAP: 32.60 -> 32.87 | Improved |
| use additive positional embeddings in cross-attention (ablation) | PViC · HICO-DET (Default, Full) | mAP: 33.59 -> 33.43 | No clear effect |
| concatenate positional embeddings (removing cross-terms) and modulate with box width/height (ablation) | PViC · HICO-DET (Default, Full) | mAP: 33.59 -> 33.91 | Improved |
| increase decoder layers beyond two (ablation) | PViC · HICO-DET (Default, Full) | mAP: 34.18 -> 34.03 | No clear effect |
| replace single-scale C5 with multiscale deformable cross-attention (ablation) | PViC · HICO-DET (Default, Full) | mAP: 34.18 -> 33.59 | Hurt |
Closest papers by meaning
- Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · similarity 0.90)
- Detecting Any Human-Object Interaction Relationship: Universal HOI Detector with Spatial Prompt Learning on Foundation Models (NeurIPS 2023 · similarity 0.90)
- Discovering Syntactic Interaction Clues for Human-Object Interaction Detection (CVPR 2024 · similarity 0.89)
- HORP: Human-Object Relation Priors Guided HOI Detection (CVPR 2025 · similarity 0.89)
- ContextHOI: Spatial Context Learning for Human-Object Interaction Detection (AAAI 2025 · similarity 0.89)
- Learning Human-Object Interaction as Groups (NeurIPS 2025 · similarity 0.89)
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.
