Research map / 3D vision / Human pose estimation
Image-Based Synthesis for Deep 3D Human Pose Estimation
Grégory Rogez, Cordelia Schmid
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 synthetic images to real images for training (ablation) | pose classifier · Human3.6M P1 | 3D pose error (mm): 97.7 -> 88.1 | Improved |
| add synthetic images to real images for training (ablation) | pose regressor · Human3.6M P1 | 3D pose error (mm): 139.6 -> 125.5 | Improved |
| train classifier on synthetic instead of real images (190,000 poses) (ablation) | pose classifier · Human3.6M P1 | 3D pose error (mm): 97.7 -> 97.2 | No clear effect |
| use classifier instead of regressor with few training images (17,000) (ablation) | pose classifier · Human3.6M P1 | 3D pose error (mm): 112.9 -> 149.7 | Hurt |
| use classifier instead of regressor with large synthetic+real data (ablation) | pose classifier · Human3.6M P1 | 3D pose error (mm): 125.5 -> 88.1 | Improved |
| train AlexNet from scratch instead of fine-tuning ImageNet model (ablation) | pose classifier · Human3.6M P1 | 3D pose error (mm): 98.3 -> 88.1 | Improved |
| render from in-the-wild 2D images (MPII+LSP) instead of H3.6M images (ablation) | pose classifier · LSP | 2D pose error (pixels): 31.1 -> 20.7 | Improved |
| render from in-the-wild 2D images (MPII+LSP) instead of H3.6M images (ablation) | pose classifier · Human3.6M | 2D pose error (pixels): 8.8 -> 17.3 | Hurt |
| use CMU 3D poses instead of H3.6M poses (190,000) (ablation) | pose classifier · LSP | 2D pose error (pixels): 20.7 -> 22.4 | Hurt |
| increase CMU poses and synthetic images from 190,000 to 2 million (ablation) | pose classifier · LSP | 2D pose error (pixels): 22.4 -> 13.8 | Improved |
| increase CMU poses and synthetic images from 190,000 to 2 million (ablation) | pose classifier · Human3.6M | 3D pose error (mm): 150.6 -> 138.0 | Improved |
| fine-tune VGG-16 instead of AlexNet (ablation) | pose classifier · LSP | 2D pose error (pixels): 13.8 -> 11.5 | Improved |
Closest papers by meaning
- MoCap-guided Data Augmentation for 3D Pose Estimation in the Wild (NeurIPS 2016 · similarity 0.97)
- 3D Human Pose Estimation = 2D Pose Estimation + Matching (CVPR 2017 · similarity 0.91)
- In the Wild Human Pose Estimation Using Explicit 2D Features and Intermediate 3D Representations (CVPR 2019 · similarity 0.90)
- A Simple yet Effective Baseline for 3D Human Pose Estimation (ICCV 2017 · similarity 0.90)
- Sim2real transfer learning for 3D human pose estimation: motion to the rescue (NeurIPS 2019 · similarity 0.90)
- 3D Human Pose Estimation in the Wild by Adversarial Learning (CVPR 2018 · similarity 0.89)
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.
