Research map / 3D vision / Human pose estimation
Optimizing Network Structure for 3D Human Pose Estimation
Hai Ci, Chunyu Wang, Xiaoxuan Ma, Yizhou 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 |
|---|---|---|---|
| replace FCN with locally connected network (3-NN) | Martinez FCN · H36M protocol #1 (SH trained on MPII 2D, pixel CS) | MPJPE: 63.21 -> 57.56 | Improved |
| replace GCN shared-weight filters with per-joint weights (LCN) (ablation) | Defferrard GCN · H36M protocol #1 (SH trained on MPII 2D, pixel CS) | MPJPE: 66.37 -> 57.56 | Improved |
| use GCN instead of FCN | FCN · H36M protocol #1 (SH trained on MPII 2D, pixel CS) | MPJPE: 63.21 -> 66.37 | Hurt |
| increase neighbourhood from 1-NN to 2-NN (ablation) | LCN · H36M protocol #1 (SH trained on MPII 2D, pixel CS) | MPJPE: 58.77 -> 57.73 | Improved |
| increase neighbourhood from 3-NN to 4-NN (ablation) | LCN · H36M protocol #1 (SH trained on MPII 2D, pixel CS) | MPJPE: 57.56 -> 58.37 | Hurt |
| learn structure matrix values on top of 2-NN (ablation) | LCN (2-NN) · H36M protocol #1 (SH trained on MPII 2D, pixel CS) | MPJPE: 57.73 -> 57.80 | No clear effect |
| learn the whole structure matrix from data instead of anatomy prior (ablation) | LCN · H36M protocol #1 (SH trained on MPII 2D, pixel CS) | MPJPE: 57.56 -> 59.22 | Hurt |
| remove pose scale using scale-normalised pixel coordinate system instead of camera CS (ablation) | Martinez FCN · H36M protocol #1 (SH trained on MPII 2D, pixel vs camera CS) | MPJPE: 67.50 -> 63.21 | Improved |
| remove pose scale using scale-normalised pixel coordinate system instead of camera CS (ablation) | LCN (3-NN) · H36M protocol #1 (SH trained on MPII 2D, pixel vs camera CS) | MPJPE: 62.95 -> 57.56 | Improved |
| LCN instead of FCN on cross-dataset test | Martinez FCN · MPI-INF-3DHP (trained on H36M unless noted) all scenes | PCK: 42.5 -> 74.0 | Improved |
| LCN instead of FCN when trained on one action | Martinez FCN · H36M cross-action training | MPJPE: about 20mm smaller | Improved |
| LCN instead of FCN with scarce training data | Martinez FCN · H36M scarce (Direction only) | MPJPE: 2k vs 5K examples to saturate | Improved |
Closest papers by meaning
- Locally Connected Network for Monocular 3D Human Pose Estimation (TPAMI 2020 · similarity 0.96)
- Graph and Temporal Convolutional Networks for 3D Multi-person Pose Estimation in Monocular Videos (AAAI 2021 · similarity 0.92)
- Modulated Graph Convolutional Network for 3D Human Pose Estimation (ICCV 2021 · similarity 0.91)
- GLA-GCN: Global-local Adaptive Graph Convolutional Network for 3D Human Pose Estimation from Monocular Video (ICCV 2023 · similarity 0.91)
- A Comprehensive Study of Weight Sharing in Graph Networks for 3D Human Pose Estimation (ECCV 2020 · similarity 0.91)
- Exploiting Spatial-Temporal Relationships for 3D Pose Estimation via Graph Convolutional Networks (ICCV 2019 · 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.
