Research map / 3D vision / Human pose estimation
Towards Annotation-Free Evaluation: KPAScore for Human Keypoint Detection
Xiaoxiao Wang, Chunxiao Li, Peng Sun, Boming Miao, Yunjian Zhang, Yao Zhu
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 |
|---|---|---|---|
| add a blue-dot visual prompt instead of numeric coordinates in the localization query (ablation) | GLM-4V-9B (zero-shot) · KPUBench precision localization | ACC (%): 49.98 -> 62.57 | Improved |
| use visual prompts instead of numeric coordinates across VLMs (ablation) | 18 VLMs zero-shot · KPUBench precision localization | ACC (%): +5.42% average | Improved |
| present numeric coordinates to VLMs | 18 VLMs zero-shot · KPUBench coordinate understanding | ACC (%) | No clear effect |
| LoRA fine-tune GLM-4V-9B on 10k visual-prompt localization samples | GLM-4V-9B · KPUBench visual prompt | ACC (%): 62.57 -> 80.15 | Improved |
| add LLaVA-V1.6-34B presence-detection first stage before localization (ablation) | fine-tuned GLM-4V-9B · KPUBench visual prompt | ACC (%): 80.15 -> 84.53 | Improved |
| increase random keypoint offset up to 50% of bbox size | KPAScore · COCO | KPAScore | Hurt |
| apply KPAScore-guided keypoint ensemble of ViTPose-H and RTMPose-l | KPAPose · COCO | OKS: 72.95 -> 73.63 | Improved |
| use ensemble KPAPose on hard actions | KPAPose · Complex scenes without ground truth | KPAScore: 63.09 -> 89.03 | Improved |
| model size scaling for presence detection | VLMs zero-shot · KPUBench presence detection | ACC (%) | No clear effect |
Closest papers by meaning
- Rethinking Keypoint Representations: Modeling Keypoints and Poses as Objects for Multi-Person Human Pose Estimation (ECCV 2022 · similarity 0.87)
- KptLLM: Unveiling the Power of Large Language Model for Keypoint Comprehension (NeurIPS 2024 · similarity 0.86)
- Incremental Object Keypoint Learning (CVPR 2025 · similarity 0.85)
- ProbPose: A Probabilistic Approach to 2D Human Pose Estimation (CVPR 2025 · similarity 0.85)
- LocLLM: Exploiting Generalizable Human Keypoint Localization via Large Language Model (CVPR 2024 · similarity 0.85)
- Back to 3D: Few-Shot 3D Keypoint Detection with Back-Projected 2D Features (CVPR 2024 · 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.
