Research map / Detection and segmentation / Open-set and few-shot detection
Improving Human-Object Interaction Detection via Phrase Learning and Label Composition
Zhimin Li, Cheng Zou, Yu Zhao, Boxun Li, Sheng Zhong
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 |
|---|---|---|---|
| Jointly train HOI detection with relational phrase learning | HOI model* · HICO-DET, HOI branch | Full mAP: 24.07 -> 29.29 | Improved |
| Use semantic-neighbor label composition rather than no augmentation (ablation) | PhraseHOI · HICO-DET | Full mAP: 27.36 -> 29.29 | Improved |
| Set the label-composition similarity threshold to 0.6 (ablation) | PhraseHOI · HICO-DET | Full mAP: 29.29 -> 27.80 | Hurt |
| Use the combined L1 distilling and triplet phrase loss instead of L1 alone (ablation) | PhraseHOI · HICO-DET | Full mAP: 28.10 -> 29.29 | Improved |
| Remove the phrase learning loss (ablation) | PhraseHOI · HICO-DET | Full mAP: 29.29 -> 24.07 | Hurt |
| Use MSE instead of the combined phrase loss (ablation) | PhraseHOI · HICO-DET | Full mAP: 29.29 -> 27.06 | Hurt |
| Use KL divergence instead of the combined phrase loss (ablation) | PhraseHOI · HICO-DET | Full mAP: 29.29 -> 26.52 | Hurt |
| Use cosine loss instead of the combined phrase loss (ablation) | PhraseHOI · HICO-DET | Full mAP: 29.29 -> 26.32 | Hurt |
| Use L1 distilling loss without triplet loss (ablation) | PhraseHOI · HICO-DET | Full mAP: 29.29 -> 28.10 | Hurt |
| Set triplet margin to 0.50 rather than removing the triplet loss (ablation) | PhraseHOI · HICO-DET | Rare mAP: 18.64 -> 22.03 | Improved |
| Set triplet margin to 1.00 (ablation) | PhraseHOI · HICO-DET | Full mAP: 29.29 -> 28.59 | Hurt |
| Use language-model encoded phrase embeddings instead of concatenated word2vec embeddings (ablation) | PhraseHOI · HICO-DET | Full mAP: 29.29 -> 22.64 | Hurt |
Closest papers by meaning
- Learning to Detect Human-Object Interactions With Knowledge (CVPR 2019 · similarity 0.91)
- Polysemy Deciphering Network for Human-Object Interaction Detection (ECCV 2020 · similarity 0.91)
- Human-Object Interaction Detection Collaborated with Large Relation-driven Diffusion Models (NeurIPS 2024 · similarity 0.90)
- Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · similarity 0.89)
- Discovering Syntactic Interaction Clues for Human-Object Interaction Detection (CVPR 2024 · similarity 0.89)
- Exploring the Potential of Large Foundation Models for Open-Vocabulary HOI Detection (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.
