Research map / 3D vision / Human pose estimation
Towards Part-aware Monocular 3D Human Pose Estimation: An Architecture Search Approach
Zerui Chen, Yan Huang, Hongyuan Yu, Bin Xue, Ke Han, Yiru Guo, Liang Wang
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 |
|---|---|---|---|
| use two intermediate nodes in the fusion cell instead of one (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 58.9 -> 53.6 | Improved |
| use two intermediate nodes instead of three (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 60.9 -> 53.6 | Improved |
| use two intermediate nodes instead of four (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 58.0 -> 53.6 | Improved |
| include dilated convolution in the search space (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 59.9 -> 53.6 | Improved |
| include separable (depth-wise) convolution in the search space (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 5.1 mm decline in performance without it (58.7 vs 53.6) | Improved |
| keep the searched part order instead of shuffling heat map order (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 57.2 -> 53.6 | Improved |
| search the whole network including fusion cells instead of backbone only (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 57.1 -> 53.6 | Improved |
| use the searched whole architecture instead of a ResNet-50 backbone | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 53.9 -> 53.6, 62% fewer parameters | Improved |
| increase number of computational cells from 10 to 15 (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 53.6 -> 50.9 | Improved |
| increase number of computational cells from 10 to 20 (ablation) | part-aware NAS pose estimator · Human3.6M Protocol 2 | MPJPE mm: 53.6 -> 47.3 | Improved |
| add supervision on intermediate feature maps | Ours, S · Human3.6M Protocol 2 | MPJPE mm: 53.6 -> 50.4 | Improved |
Closest papers by meaning
- Learning a Robust Part-Aware Monocular 3D Human Pose Estimator via Neural Architecture Search (IJCV 2021 · similarity 0.93)
- Learning to Fuse 2D and 3D Image Cues for Monocular Body Pose Estimation (ICCV 2017 · similarity 0.91)
- A Simple yet Effective Baseline for 3D Human Pose Estimation (ICCV 2017 · similarity 0.91)
- 3D Human Pose Estimation From a Single Image via Distance Matrix Regression (CVPR 2017 · similarity 0.90)
- Generating Multiple Hypotheses for 3D Human Pose Estimation With Mixture Density Network (CVPR 2019 · similarity 0.90)
- Sparseness Meets Deepness: 3D Human Pose Estimation From Monocular Video (CVPR 2016 · similarity 0.90)
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.
