Research map / 3D vision / Human pose estimation
HAP: Structure-Aware Masked Image Modeling for Human-Centric Perception
Junkun Yuan, Xinyu Zhang, Hao Zhou, Jian Wang, Zhongwei Qiu, Zhiyin Shao, Shaofeng Zhang, Sifan Long, Kun Kuang, Kun Yao, Junyu Han, Errui Ding and 3 more
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 |
|---|---|---|---|
| Use human structure settings: high scaling ratio, intermediate masking ratio, and block-wise masking (ablation) | MAE pre-training on LUPerson · MSMT17 | mAP: 68.9 -> 74.1 | Improved |
| Add human-part-guided mask sampling (ablation) | HAP-0 · MSMT17 | mAP: 74.1 -> 75.1 | Improved |
| Add structure-invariant alignment loss (ablation) | HAP-0 · MSMT17 | mAP: 74.1 -> 75.6 | Improved |
| Use attention-map guidance instead of part guidance (ablation) | HAP mask sampling · MSMT17 | mAP: 75.6 -> 71.7 | Hurt |
| Guide masking with keypoints rather than whole human parts (ablation) | HAP mask sampling · MSMT17 | mAP: 75.6 -> 75.9 | No clear effect |
| Select a fixed 3 parts instead of sampling the number of parts uniformly from 0 to 6 (ablation) | HAP mask sampling · MSMT17 | mAP: 76.4 -> 75.4 | Hurt |
| Select a fixed 6 parts instead of sampling the number of parts uniformly from 0 to 6 (ablation) | HAP mask sampling · MSMT17 | mAP: 76.4 -> 74.4 | Hurt |
| Predict keypoints instead of reconstructing image pixels (ablation) | HAP decoder objective · MSMT17 | mAP: 50.3 vs. 72.2 | Hurt |
| Use OpenPose instead of ViTPose to extract keypoints (ablation) | HAP pre-training · MSMT17 | mAP: 72.0 vs. 72.2 | No clear effect |
| Set alignment-loss weight γ to 0 (ablation) | HAP · MSMT17 | mAP: 76.4 -> 75.1 | Hurt |
| Align a masked view with a global unmasked view instead of two masked views (ablation) | HAP structure-invariant alignment · MSMT17 | mAP: 76.4 -> 74.5 | Hurt |
| Use SimSiam-style alignment without negative samples (ablation) | HAP alignment loss · MSMT17 | mAP: 76.4 vs. 76.1 | No clear effect |
Closest papers by meaning
- Human De-Occlusion: Invisible Perception and Recovery for Humans (CVPR 2021 · similarity 0.87)
- UniHCP: A Unified Model for Human-Centric Perceptions (CVPR 2023 · similarity 0.87)
- Not Too Generative, Not Too Discriminative: The Human Alignment Sweet Spot (NeurIPS 2026 · similarity 0.85)
- Cross-view and Cross-pose Completion for 3D Human Understanding (CVPR 2024 · similarity 0.85)
- Evaluating Multiview Object Consistency in Humans and Image Models (NeurIPS 2024 · similarity 0.85)
- ODHSR: Online Dense 3D Reconstruction of Humans and Scenes from Monocular Videos (CVPR 2025 · similarity 0.85)
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.
