Research map / 3D vision / Human pose estimation
Occlusion-Resilient Category-Agnostic Pose Estimation with Conditional Flow Matching
Jiyong Rao, Shengjie Zhao, Yu Wang, Hao Deng
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 heatmap-guided initialization over center (query-centered, 30 steps) initialization (ablation) | FlowCape · Occ80 Split 1 | PCK@0.2: 84.77 -> 86.72 | Improved |
| add confidence weighting to heatmap-guided initialization (ablation) | FlowCape · Occ80 Split 1 | PCK@0.2: 86.72 -> 86.71 | No clear effect |
| run 30 ODE steps instead of 20 for center-initialized flow (ablation) | FlowCape · Occ80 Split 1 | PCK@0.2: 83.13 -> 84.77 | Improved |
| add H-Init, flow matching losses and auxiliary losses on top of ODE-only (ablation) | FlowCape · Occ80 Split 1 | PCK@0.2: 72.81 -> 86.72 | Improved |
| add Lrfm and Lflow flow-matching supervision (ablation) | FlowCape · Occ80 Split 1 | PCK@0.2: +0.9% | Improved |
| use full FlowCape instead of CapeX baseline (ablation) | FlowCape · Occ80 Split 1 (SwinV2-Tiny) | PCK@0.2: 85.68 -> 86.72 | Improved |
| increase pose head depth from 1 to 3 layers (ablation) | Riemannian Pose Head · Occ80 Split 1 | PCK@0.2: 83.51 -> 86.72 | Improved |
| increase pose head depth from 3 to 4 layers (ablation) | Riemannian Pose Head · Occ80 Split 1 | PCK@0.2: 86.72 -> 84.96 | Hurt |
| increase hidden dimension from 256 to 384 (ablation) | Riemannian Pose Head · Occ80 Split 1 | PCK@0.2: 86.72 -> 86.77 | No clear effect |
| increase hidden dimension from 128 to 256 (ablation) | Riemannian Pose Head · Occ80 Split 1 | PCK@0.2: 85.61 -> 86.72 | Improved |
| lower lambda_traj from 1.0 to 10^-3 (ablation) | FlowCape · Occ80 Split 1 | PCK@0.2: 86.72 -> 84.79 | Hurt |
| use FlowCape instead of CapeX with SwinV2-Small | FlowCape · Occ80 (avg of S1-S3) | PCK@0.2: 88.07 -> 89.12 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- CapeX — CapeX: Category-Agnostic Pose Estimation from Textual Point Explanation (ICLR 2025) · better on 4 of 4 shared results
- GenCape — GenCape: Structure-Inductive Generative Modeling for Category-Agnostic Pose Estimation (ICLR 2026) · better on 3 of 3 shared results
- SCAPE — SCAPE: A Simple and Strong Category-Agnostic Pose Estimator (ECCV 2024) · 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
- Matching Is Not Enough: A Two-Stage Framework for Category-Agnostic Pose Estimation (CVPR 2023 · similarity 0.91)
- A Graph-Based Approach for Category-Agnostic Pose Estimation (ECCV 2024 · similarity 0.90)
- SCAPE: A Simple and Strong Category-Agnostic Pose Estimator (ECCV 2024 · similarity 0.90)
- CapeX: Category-Agnostic Pose Estimation from Textual Point Explanation (ICLR 2025 · similarity 0.89)
- ESCAPE: Encoding Super-keypoints for Category-Agnostic Pose Estimation (CVPR 2024 · similarity 0.88)
- Meta-Point Learning and Refining for Category-Agnostic Pose Estimation (CVPR 2024 · similarity 0.88)
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.
