Research map / Detection and segmentation / Open-set and few-shot detection
Reformulating HOI Detection As Adaptive Set Prediction
Mingfei Chen, Yue Liao, Si Liu, Zhiyuan Chen, Fei Wang, Chen Qian
Check 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 |
|---|---|---|---|
| reformulate HOI detection as adaptive set prediction with learned interaction queries and matched supervision | AS-Net · HICO-DET test, Default Full | mAP: 21.94 -> 28.87 | Improved |
| combine vector matching with semantic embedding matching (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 28.56 -> 28.87; 28.65 -> 28.87 | Improved |
| set semantic embedding dimension to 4 instead of 8 (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 28.87 -> 28.21 | Hurt |
| set semantic embedding dimension to 16 or 32 instead of 8 (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 28.87 -> 28.36; 28.87 -> 28.70 | Hurt |
| set embedding loss weight to 0.05 or 0.5 instead of 0.1 (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 28.87 -> 28.31; 28.87 -> 27.84 | Hurt |
| replace parallel instance and interaction branches with a single branch (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 28.87 -> 25.91 | Hurt |
| add an interaction branch to the single-branch model (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 25.91 -> 27.52 | Improved |
| add instance-aware attention to the basic model without embeddings (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 27.52 -> 27.96 | Improved |
| add semantic embeddings to the basic model without instance-aware attention (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 27.52 -> 27.75 | Improved |
| add instance-aware attention to the model with semantic embeddings (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 27.75 -> 28.87 | Improved |
| use three instance-aware attention modules rather than six with embeddings and a 6-layer interaction decoder (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 28.87 -> 28.39 | Hurt |
| reduce both the instance-aware attention modules and interaction decoder layers to three (ablation) | AS-Net · HICO-DET test, Default Full | mAP: 28.87 -> 28.63 | Hurt |
Closest papers by meaning
- Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · similarity 0.91)
- Efficient Adaptive Human-Object Interaction Detection with Concept-guided Memory (ICCV 2023 · similarity 0.91)
- Learning Human-Object Interaction Detection Using Interaction Points (CVPR 2020 · similarity 0.91)
- Discovering Human Interactions With Large-Vocabulary Objects via Query and Multi-Scale Detection (ICCV 2021 · similarity 0.91)
- Toward Open-Set Human Object Interaction Detection (AAAI 2024 · similarity 0.89)
- Open-World Human-Object Interaction Detection via Multi-modal Prompts (CVPR 2024 · 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.
