Research map / 3D vision / Human pose estimation
Lifting 2D Human Pose to 3D with Domain Adapted 3D Body Concept
Qiang Nie, Ziwei Liu, Yunhui Liu
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 the learned 3D body concept (domain adaptation) to the linear lifting network (ablation) | linear lifting network · H3.6M (SH 2D, 5%S1, supervised) | MPJPE: improved by 3.73% | Improved |
| add the 3D body concept with GT 2D poses (ablation) | linear lifting network · H3.6M (GT 2D, 5%S1, supervised) | MPJPE: 7.86% in supervised mode | Improved |
| add the 3D body concept, average over data sizes (ablation) | linear lifting network · H3.6M (SH 2D) | MPJPE: 4.4% better on average | Improved |
| add the 3D body concept, average over data sizes (ablation) | linear lifting network · H3.6M (GT 2D) | MPJPE: 10.2% on average | Improved |
| add unlabeled 2D poses (semi-supervised mode) (ablation) | Ours · H3.6M (SH 2D, 5%S1) | MPJPE: 8.61% increase over supervised | Improved |
| add unlabeled 2D poses (semi-supervised mode) (ablation) | Ours · H3.6M (SH 2D, 50%S1 and S1) | MPJPE: down 0.11% and 0.44% | Hurt |
| add transductive inference using test data as unlabeled (ablation) | Ours · H3.6M All training data | MPJPE: 0.08%, 0.09% and 4.76% on SH, HRN and GT | Improved |
| add unlabeled data when training on all data with GT 2D (ablation) | Ours semi. · H3.6M All, GT 2D | MPJPE: 33.7 -> 35.5 | Hurt |
| add 3D body concept in cross-action training on A1 (ablation) | Ours · H3.6M cross-action (SH) | MPJPE: 135.4 -> 124.9 | Improved |
| add 3D body concept in cross-action training on A1-8 (ablation) | Ours · H3.6M cross-action (SH) | MPJPE: 88.75 -> 75.56 | Improved |
| add unlabeled data in cross-action A1-8 (ablation) | Ours semi. · H3.6M cross-action (SH) | MPJPE: 75.56 -> 79.79 | Hurt |
| add synthetic extra data E to HRN-based training of S1 | Ours sup. · H3.6M S1 (HRNet 2D) | MPJPE: 64.3 -> 59.9 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- RepNet — RepNet: Weakly Supervised Training of an Adversarial Reprojection Network for 3D Human Pose Estimation (CVPR 2019) · better on 3 of 3 shared results
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
- Lifting by Image – Leveraging Image Cues for Accurate 3D Human Pose Estimation (AAAI 2024 · similarity 0.91)
- Unsupervised 3D Pose Estimation With Geometric Self-Supervision (CVPR 2019 · similarity 0.90)
- Unite the People: Closing the Loop Between 3D and 2D Human Representations (CVPR 2017 · similarity 0.90)
- 3D Human Pose Machines with Self-supervised Learning (TPAMI 2019 · similarity 0.89)
- Keep It SMPL: Automatic Estimation of 3D Human Pose and Shape from a Single Image (ECCV 2016 · similarity 0.89)
- PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space (NeurIPS 2025 · 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.
