Research map / 3D vision / Human pose estimation
Cascaded Deep Monocular 3D Human Pose Estimation With Evolutionary Training Data
Shichao Li, Lei Ke, Kevin Pratama, Yu-Wing Tai, Chi-Keung Tang, Kwang-Ting Cheng
Check 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 |
|---|---|---|---|
| Evolve the S1 training poses and train TAG-Net on the augmented pairs (ablation) | 2D-to-3D pose estimation · Human3.6M, P1 | MPJPE: 71.5 -> 60.8 | Improved |
| Add cascade to the baseline (ablation) | 2D-to-3D pose estimation with S1 · Human3.6M, P1 | MPJPE: 71.5 -> 70.1 | Improved |
| Add data evolution to the cascaded model (ablation) | 2D-to-3D pose estimation with S1 · Human3.6M, P1* | MPJPE: 64.5 -> 50.5 | Improved |
| Add cascade to the baseline (ablation) | 2D-to-3D pose estimation with S15678 · Human3.6M, P1 | MPJPE: 54.3 -> 52.1 | Improved |
| Add data evolution to the cascaded model (ablation) | 2D-to-3D pose estimation with S15678 · Human3.6M, P1* | MPJPE: 42.9 -> 34.5 | Improved |
| Add heat-map up-sampling (ablation) | HRN 2D key-point detector · Human3.6M test | 2D key-point localization error (pixels): 4.98 -> 4.64 | Improved |
| Add soft-argmax (ablation) | HRN with heat-map up-sampling · Human3.6M test | 2D key-point localization error (pixels): 4.64 -> 4.36 | Improved |
| Add mutation alongside crossover during data evolution (ablation) | Evolved pose training data · Human3.6M, P1* | MPJPE | Improved |
| Concatenate current pose estimates to later cascade inputs (ablation) | Cascaded 2D-to-3D model · Human3.6M | MPJPE | No clear effect |
| Evolve the H36M training distribution and train on 8 million 2D-3D pairs | 2D-to-3D pose estimation · MPI-INF-3DHP, cross-dataset | AUC: 40.8 -> 46.1 | Improved |
| Evolve the H36M training distribution and train on 8 million 2D-3D pairs | 2D-to-3D pose estimation · MPI-INF-3DHP, cross-dataset | MPJPE: 117.6 -> 99.7 | Improved |
| Use TAG-Net with evolved S15678 data rather than the strongest reported baseline | Single-image 3D pose estimation · Human3.6M, S15678, P1 | MPJPE: 54.4 -> 50.9 | Improved |
Closest papers by meaning
- Monocular 3D Pose Estimation via Pose Grammar and Data Augmentation (TPAMI 2021 · similarity 0.90)
- In the Wild Human Pose Estimation Using Explicit 2D Features and Intermediate 3D Representations (CVPR 2019 · similarity 0.89)
- Learning to Fuse 2D and 3D Image Cues for Monocular Body Pose Estimation (ICCV 2017 · similarity 0.89)
- A Simple yet Effective Baseline for 3D Human Pose Estimation (ICCV 2017 · similarity 0.89)
- Towards Part-aware Monocular 3D Human Pose Estimation: An Architecture Search Approach (ECCV 2020 · similarity 0.89)
- MonoCap: Monocular Human Motion Capture using a CNN Coupled with a Geometric Prior (TPAMI 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.
