Research map / 3D vision / Human pose estimation
Ordinal Depth Supervision for 3D Human Pose Estimation
Georgios Pavlakos, Xiaowei Zhou, Kostas Daniilidis
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 2D keypoint examples to Human3.6M training (ablation) | 3D pose ConvNet · Human3.6M | mean per joint error (mm): 71.9 -> 66.6 | Improved |
| Add ordinal depth supervision to 2D-keypoint training examples (ablation) | 3D pose ConvNet · Human3.6M | mean per joint error (mm): 66.6 -> 62.1 | Improved |
| Add the reconstruction component (ablation) | 3D pose ConvNet with 2D keypoint and ordinal supervision · Human3.6M | mean per joint error (mm): 62.1 -> 56.2 | Improved |
| Add ordinal depth supervision to in-the-wild 2D-keypoint examples (ablation) | 3D pose ConvNet trained on Human3.6M · MPI-INF-3DHP | PCK3D: 44.3 -> 71.9 | Improved |
| Use ordinal supervision rather than 3D-ground-truth direct regression (ablation) | Depth prediction ConvNet · Human3.6M | mean per joint error (mm): 80.23 -> 84.24 | Hurt |
| Use weak ordinal-depth and 2D-keypoint supervision rather than full 3D supervision (ablation) | Coordinate regression ConvNet · Human3.6M | mean per joint error (mm): 112.41 -> 115.08 | Hurt |
| Use ordinal supervision rather than full 3D supervision (ablation) | Volume regression ConvNet with one hourglass · Human3.6M | mean per joint error (mm): 85.82 -> 89.93 | Hurt |
| Use ordinal supervision rather than full 3D supervision (ablation) | Volume regression ConvNet with two hourglasses · Human3.6M | mean per joint error (mm): 69.77 -> 79.03 | Hurt |
| Add in-the-wild 2D-keypoint and ordinal-depth examples (ablation) | 3D pose ConvNet trained on Human3.6M · MPI-INF-3DHP | AUC: 6.3 -> 19.8 -> 35.3 | Improved |
Closest papers by meaning
- HMOR: Hierarchical Multi-Person Ordinal Relations for Monocular Multi-Person 3D Pose Estimation (ECCV 2020 · similarity 0.90)
- A Simple yet Effective Baseline for 3D Human Pose Estimation (ICCV 2017 · similarity 0.90)
- Monocular 3D Human Pose Estimation by Generation and Ordinal Ranking (ICCV 2019 · similarity 0.90)
- Coarse-To-Fine Volumetric Prediction for Single-Image 3D Human Pose (CVPR 2017 · similarity 0.88)
- 3D Human Pose Estimation From a Single Image via Distance Matrix Regression (CVPR 2017 · similarity 0.88)
- PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space (NeurIPS 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.
