Research map / 3D vision / Human pose estimation
HigherHRNet: Scale-Aware Representation Learning for Bottom-Up Human Pose Estimation
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 |
|---|---|---|---|
| Add a deconvolution module to predict higher-resolution heatmaps (ablation) | HRNet bottom-up pose estimation · COCO2017 val | AP: 64.4 -> 66.0 | Helped |
| Add feature concatenation of features and predicted heatmaps (ablation) | HigherHRNet with multi-resolution supervision · COCO2017 val | AP: 66.0 -> 66.3 | Helped |
| Aggregate heatmaps from all resolutions at inference (ablation) | HigherHRNet with feature concatenation · COCO2017 val | AP: 66.3 -> 66.9 | Helped |
| Add 4 residual blocks to the deconvolution module (ablation) | HigherHRNet with heatmap aggregation · COCO2017 val | AP: 66.9 -> 67.1 | Helped |
| Use two deconvolution modules instead of one (ablation) | HigherHRNet · COCO2017 val | AP: 66.9 -> 66.5 | Hurt |
| Increase training image size from 512 to 640 (ablation) | HigherHRNet · COCO2017 val | AP: 67.1 -> 68.5 | Helped |
| Increase training image size from 640 to 768 (ablation) | HigherHRNet · COCO2017 val | AP: 68.5 -> 68.5 | No clear effect |
| Replace HRNet-W32 backbone with HRNet-W40 (ablation) | HigherHRNet · COCO2017 val | AP: 68.5 -> 69.2 | Helped |
| Replace HRNet-W40 backbone with HRNet-W48 (ablation) | HigherHRNet · COCO2017 val | AP: 69.2 -> 69.9 | Helped |
| Add one deconvolution module to HRNet to produce stride-2 heatmaps (ablation) | HRNet bottom-up pose estimation · COCO2017 val, medium-person AP | AP^M: 57.1 -> 61.0 | Helped |
| Use higher-resolution heatmaps without multi-resolution aggregation (ablation) | HigherHRNet · COCO2017 val, large-person AP | AP^L: 75.6 -> 74.0 | Hurt |
| Use multi-resolution heatmap aggregation rather than only the highest-resolution heatmap (ablation) | HigherHRNet with feature concatenation · COCO2017 val, large-person AP | AP^L: +1.7 AP | Helped |
Closest papers by meaning
- Deep High-Resolution Representation Learning for Human Pose Estimation (CVPR 2019 · similarity 0.89)
- Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation (AAAI 2020 · similarity 0.88)
- SIMPLE: SIngle-network with Mimicking and Point Learning for Bottom-up Human Pose Estimation (AAAI 2021 · similarity 0.88)
- Rethinking the Heatmap Regression for Bottom-Up Human Pose Estimation (CVPR 2021 · similarity 0.87)
- SHaRPose: Sparse High-Resolution Representation for Human Pose Estimation (AAAI 2024 · similarity 0.85)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 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.