Research map / Detection and segmentation / CNN features
MSTR: Multi-Scale Transformer for End-to-End Human-Object Interaction Detection
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Replace QPIC attention with deformable attention (ablation) | QPIC · HICO-DET test, Default Full | mAP: 29.07 -> 25.53 | Hurt |
| Add Dual-Entity attention to single-scale deformable attention (ablation) | SS-Baseline · HICO-DET test, Default Full | mAP: 25.53 -> 27.06 (+1.53p) | Helped |
| Add Entity-conditioned Context attention after Dual-Entity attention (ablation) | SS-Baseline + DE · HICO-DET test, Default Full | mAP: 27.06 -> 27.70 (+0.64p) | Helped |
| Use multi-scale features instead of single-scale features with deformable attention (ablation) | Deformable-attention baseline · HICO-DET test, Default Full | mAP: 25.53 -> 27.52 | Helped |
| Add Dual-Entity attention to multi-scale deformable attention (ablation) | MS-Baseline · HICO-DET test, Default Full | mAP: 27.52 -> 28.30 (+0.78p) | Helped |
| Add Entity-conditioned Context attention after Dual-Entity attention (ablation) | MS-Baseline + DE · HICO-DET test, Default Full | mAP: 28.30 -> 30.14 (+1.84p) | Helped |
| Use multi-scale rather than single-scale features with DE and EC (ablation) | Deformable-attention baseline with DE and EC · HICO-DET test, Default Full | mAP: 27.70 -> 30.14 | Helped |
| Merge separately self-attended semantic outputs instead of merging self-attention inputs (ablation) | MSTR decoder · HICO-DET test, Default Full | mAP: 30.14 -> 31.17 (+1.03p) | Helped |
| Use separate human and object sampling points and attention (ablation) | QPIC with deformable attention · HICO-DET test, Default Full | mAP: 27.52 -> 28.30 | Helped |
| Use separate decoder stacks for human and object semantics instead of a shared self-attention stream (ablation) | Dual-Entity attention decoder · HICO-DET test, Default Full | mAP: 28.30 -> 28.15 | Hurt |
| Condition context sampling references on human and object reference points (ablation) | Context attention with Dual-Entity attention · HICO-DET test, Default Full | mAP: 29.36 -> 30.14 (+0.78p) | Helped |
| Use multi-scale MSTR instead of single-scale QPIC | HOI detection · HICO-DET, human-object relative-area groups | AP: 34.10 -> 37.11; 30.57 -> 31.68; 25.22 -> 27.07 | Helped |
Closest papers by meaning
- End-to-End Human Object Interaction Detection With HOI Transformer (CVPR 2021 · similarity 0.92)
- HOTR: End-to-End Human-Object Interaction Detection With Transformers (CVPR 2021 · similarity 0.92)
- Exploring Pose-Aware Human-Object Interaction via Hybrid Learning (CVPR 2024 · similarity 0.88)
- Orchestrating the Symphony of Prompt Distribution Learning for Human-Object Interaction Detection (AAAI 2025 · similarity 0.88)
- Toward Open-Set Human Object Interaction Detection (AAAI 2024 · similarity 0.88)
- HOIMamba: Efficient Mamba-based Disentangled Progressive Learning for HOI Detection (AAAI 2025 · 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.