Research map / 3D vision / Human pose estimation
The Center of Attention: Center-Keypoint Grouping via Attention for Multi-Person Pose Estimation
Guillem Brasó, Nikita Kister, Laura Leal-Taixé
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 |
|---|---|---|---|
| replace associative-embedding grouping with attention-based CenterGroup grouping (no transformer, same-type) (ablation) | CenterGroup (HRNet32) · COCO val2017 | AP: 67.1 -> 67.5 | Improved |
| replace offsets + keypoint matching with attention grouping (ablation) | CenterGroup (HRNet32) · COCO val2017 | AP: 65.3 -> 67.5 | Improved |
| make attention type-agnostic (attend over all keypoint types) (ablation) | CenterGroup w/o transformer · COCO val2017 | AP: 67.5 -> 67.5 | No clear effect |
| add learned type encoding to type-agnostic attention (ablation) | CenterGroup w/o transformer · COCO val2017 | AP: 67.5 -> 67.9 | Improved |
| add keypoint and center transformer encoder (ablation) | CenterGroup · COCO val2017 | AP: 67.9 -> 68.4 | Improved |
| add positional encodings to transformer input (ablation) | CenterGroup · COCO val2017 | AP: 68.4 -> 68.6 | Improved |
| score poses with center classification instead of heatmap score (ablation) | CenterGroup · COCO val2017 | AP: 66.5 -> 67.1 | Improved |
| add keypoint visibility prediction with a simple MLP (ablation) | CenterGroup · COCO val2017 | AP: 67.1 -> 68.2 | Improved |
| predict visibility with attention-based head instead of MLP (ablation) | CenterGroup · COCO val2017 | AP: 68.2 -> 68.6 | Improved |
| CenterGroup grouping instead of AE grouping (HigherHRNet), speed | HigherHRNet · COCO val2017 | grouping time (ms): 327 -> 52 | Improved |
| CenterGroup grouping instead of HigherHRNet on CrowdPose | HrHRNet-W48 · CrowdPose test | AP: 65.9 -> 67.6 | Improved |
| CenterGroup grouping instead of HigherHRNet on COCO test-dev | HrHRNet-W32 · COCO test-dev2017 (single-scale) | AP: 66.4 -> 67.6 | Improved |
Closest papers by meaning
- Group Pose: A Simple Baseline for End-to-End Multi-Person Pose Estimation (ICCV 2023 · similarity 0.91)
- Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation (AAAI 2020 · similarity 0.89)
- Keypoint Communities (ICCV 2021 · similarity 0.88)
- DeepCut: Joint Subset Partition and Labeling for Multi Person Pose Estimation (CVPR 2016 · similarity 0.87)
- QueryPose: Sparse Multi-Person Pose Regression via Spatial-Aware Part-Level Query (NeurIPS 2022 · similarity 0.87)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.87)
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.
