Research map / 3D vision / Human pose estimation
SCAPE: A Simple and Strong Category-Agnostic Pose Estimator
Yujia Liang, Zixuan Ye, Wenze Liu, Hao Lu
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 explicit similarity-map matching with implicit attention matching and direct coordinate regression (ablation) | SCAPE (w/o GKP, KAR) · MP-100 1-shot split1 | PCK: 86.3 -> 89.1 | Improved |
| replace DETR decoder with full self-attention encoder (ablation) | SCAPE · MP-100 1-shot split1 | PCK: 88.6 -> 89.1 | Improved |
| unshare Q/K projection weights between keypoints and query image (ablation) | SCAPE · MP-100 1-shot split1 | PCK: 89.1 -> 89.8 | Improved |
| add Global Keypoint feature Perceptor (GKP) (ablation) | SCAPE · MP-100 1-shot split1 | PCK: 89.8 -> 90.8 | Improved |
| add Keypoint Attention Refiner (KAR) (ablation) | SCAPE · MP-100 1-shot split1 | PCK: 89.8 -> 91.2 | Improved |
| use 4 attention filters instead of 1 in KAR (ablation) | KAR · MP-100 1-shot split1 | PCK: 90.5 -> 91.2 | Improved |
| increase hidden dimension of a single attention filter to match parameters (ablation) | KAR · MP-100 1-shot split1 | PCK: 90.5 -> 90.3 | Hurt |
| use 2 GKP + 4 interactor layers instead of all 6 interactor (ablation) | SCAPE · MP-100 1-shot split1 | PCK: 91.2 -> 91.6 | Improved |
| add support image into the feature interaction stage (ablation) | SCAPE without GKP · MP-100 1-shot split1 | PCK: 89.8 -> 89.3 | Hurt |
| use DINOv2 ViT-B backbone instead of ResNet-50 | SCAPE · MP-100 1-shot mean | PCK: 87.55 -> 91.95 | Improved |
| apply GKP and KAR to CapeFormer | CapeFormer · MP-100 split1 | PCK: +1.1 PCK | Improved |
| use generalist models (Painter) instead of CAPE-specific model | MP-100 keypoint localization · MP-100 split1 | PCK: 91.6 -> 23.9 | Hurt |
Later papers that compared with this method
3 later accepted papers report SCAPE as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- GenCape: Structure-Inductive Generative Modeling for Category-Agnostic Pose Estimation (ICLR 2026 · better on 2 of 2 shared results)
- EdgeCape: Edge Weight Prediction For Category-Agnostic Pose Estimation (ICLR 2026 · better on 2 of 2 shared results)
- Occlusion-Resilient Category-Agnostic Pose Estimation with Conditional Flow Matching (ECCV 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CapeX: Category-Agnostic Pose Estimation from Textual Point Explanation (ICLR 2025 · similarity 0.93)
- ESCAPE: Encoding Super-keypoints for Category-Agnostic Pose Estimation (CVPR 2024 · similarity 0.92)
- Meta-Point Learning and Refining for Category-Agnostic Pose Estimation (CVPR 2024 · similarity 0.91)
- SecondPose: SE(3)-Consistent Dual-Stream Feature Fusion for Category-Level Pose Estimation (CVPR 2024 · similarity 0.88)
- SchemaPose: RGB-Based Category-Level Pose Estimation with Parametric Category Schema (NeurIPS 2026 · similarity 0.88)
- Universal Features Guided Zero-Shot Category-Level Object Pose Estimation (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.
