Research map / 3D vision / Human pose estimation
AutoLink: Self-supervised Learning of Human Skeletons and Object Outlines by Linking Keypoints
Xingzhe He, Bastian Wandt, Helge Rhodin
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 learned graph edges (vs keypoints only) (ablation) | AutoLink · Taichi | MAE: 722 -> 316 | Improved |
| add learned graph edges (vs keypoints only) (ablation) | AutoLink · DeepFashion | PCK: 52.9 -> 65.8 | Improved |
| add learned graph edges (vs keypoints only) (ablation) | AutoLink · Human3.6m | error: 3.58 -> 2.76 | Improved |
| use single-channel max edge map instead of edge-specific channels (ablation) | AutoLink · Human3.6m | error: 3.83 -> 2.76 | Improved |
| make alpha learnable instead of fixed (ablation) | AutoLink · CelebA Wild | error: 6.39 -> 5.24 | Improved |
| learn edge thickness instead of fixed thickness (ablation) | AutoLink · Taichi | MAE: 316 -> 311 (but worse on others) | Mixed |
| increase number of keypoints (ablation) | AutoLink · CelebA Wild | error | Improved |
| vary edge thickness over orders of magnitude (ablation) | AutoLink · several | error | No clear effect |
| increase mask ratio from 10% to 80% (16x16 patches) (ablation) | AutoLink · CelebA Wild | error: 49.2 -> 5.24 | Improved |
| use too small (4x4) or too large (64x64) mask patches (ablation) | AutoLink · Human3.6m | error: 2.76 -> 4.08 (64x64) | Hurt |
| tune edge thickness per dataset | AutoLink · Taichi | MAE: 337.50 -> 316.10 | Improved |
Closest papers by meaning
- Self-Supervised 3D Human Pose Estimation via Part Guided Novel Image Synthesis (CVPR 2020 · similarity 0.87)
- Self-Supervised Neural Articulated Shape and Appearance Models (CVPR 2022 · similarity 0.87)
- SkeletonNet: A Topology-Preserving Solution for Learning Mesh Reconstruction of Object Surfaces From RGB Images (TPAMI 2021 · similarity 0.86)
- Diffusion-Driven Self-Supervised Learning for Shape Reconstruction and Pose Estimation (TPAMI 2025 · similarity 0.86)
- SAMoR: Motion Modelling for Articulated Objects of Any Skeleton and Topology (NeurIPS 2026 · similarity 0.86)
- Unsupervised 3D Structure Inference from Category-Specific Image Collections (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.
