Research map / 3D vision / Human pose estimation
A Simple yet Effective Baseline for 3D Human Pose Estimation
Julieta Martinez, Rayat Hossain, Javier Romero, James J. Little
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 batch normalization (ablation) | 2D-to-3D MLP baseline · Human3.6M protocol 1 | MPJPE (mm): 88.5 -> 67.5 | Improved |
| add dropout (ablation) | 2D-to-3D MLP baseline · Human3.6M protocol 1 | MPJPE (mm): 71.4 -> 67.5 | Improved |
| add both batch normalization and dropout (ablation) | 2D-to-3D MLP baseline · Human3.6M protocol 1 | MPJPE (mm): 76.0 -> 67.5 | Improved |
| add residual connections (ablation) | 2D-to-3D MLP baseline · Human3.6M protocol 1 | MPJPE (mm): 75.8 -> 67.5 | Improved |
| predict 3D pose in camera coordinates (ablation) | 2D-to-3D MLP baseline · Human3.6M protocol 1 | MPJPE (mm): 101.1 -> 67.5 | Improved |
| use 2 residual blocks instead of 1 (ablation) | 2D-to-3D MLP baseline · Human3.6M protocol 1 | MPJPE (mm): 74.2 -> 67.5 | Improved |
| increase residual blocks from 2 to 4 (ablation) | 2D-to-3D MLP baseline · Human3.6M protocol 1 | MPJPE (mm): 67.5 -> 69.3 | Hurt |
| increase residual blocks from 2 to 8 (ablation) | 2D-to-3D MLP baseline · Human3.6M protocol 1 | MPJPE (mm): 67.5 -> 69.7 | Hurt |
| fine-tune the stacked hourglass 2D detector on Human3.6M | Ours (SH detections) (MA) · Human3.6M protocol 1 | MPJPE (mm): 67.5 -> 62.9 | Improved |
| train one model for all actions instead of per action | Ours (SH detections) · Human3.6M protocol 1 | MPJPE (mm): 73.6 -> 67.5 | Improved |
| use ground-truth 2D instead of stacked hourglass detections | Ours (MA) · Human3.6M protocol 1 | MPJPE (mm): 67.5 -> 45.5 | Improved |
| fine-tune the detector on Human3.6M | Ours (MA, 17 joints) · Human3.6M protocol 2 | MPJPE (mm): 52.5 -> 47.7 | Improved |
Closest papers by meaning
- Lifting From the Deep: Convolutional 3D Pose Estimation From a Single Image (CVPR 2017 · similarity 0.94)
- Coarse-To-Fine Volumetric Prediction for Single-Image 3D Human Pose (CVPR 2017 · similarity 0.94)
- Exploiting temporal information for 3D human pose estimation (ECCV 2018 · similarity 0.94)
- 3D Human Pose Estimation = 2D Pose Estimation + Matching (CVPR 2017 · similarity 0.94)
- 3D Human Pose Estimation From a Single Image via Distance Matrix Regression (CVPR 2017 · similarity 0.93)
- PandaPose: 3D Human Pose Lifting from a Single Image via Propagating 2D Pose Prior to 3D Anchor Space (NeurIPS 2025 · similarity 0.91)
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.
