Research map / 3D vision / Human pose estimation
Multi-task Deep Learning for Real-Time 3D Human Pose Estimation and Action Recognition
Diogo Carbonera Luvizon, David Picard, Hedi Tabia
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 action recognition (multi-task training) to pose estimation (ablation) | Multi-task network · Human3.6M | MPJPE (mm): 49.5 -> 48.6 | Improved |
| add pose supervision to action-only training (ablation) | Multi-task network · Penn Action | accuracy (%): 87.5 -> 97.4 | Improved |
| add MPII data for pose supervision (ablation) | Multi-task network · Penn Action | accuracy (%): 97.4 -> 98.2 | Improved |
| add multi-task training (pose supervision) (ablation) | Multi-task network · NTU RGB+D (cross-subject) | accuracy (%): 88.0 -> 89.9 | Improved |
| supervise all prediction blocks instead of only the last (ablation) | Multi-PB network (10M) · MPII val | PCKh (%): 84.2 -> 85.1 | Improved |
| supervise all prediction blocks instead of only the last (ablation) | Multi-PB network (10M) · Penn Action | accuracy (%): 97.5 -> 98.1 | Improved |
| scale to full Multi-PB network (P=8, L=4, 26M) (ablation) | Multi-PB network · MPII val | PCKh (%): 85.1 -> 88.3 | Improved |
| add appearance features to pose features (ablation) | Action recognition part · Penn Action | accuracy (%): 97.4 -> 98.1 | Improved |
| add pose features to appearance features (ablation) | Action recognition part · Penn Action | accuracy (%): 97.9 -> 98.1 | Improved |
| add decoupled action poses (ablation) | Combined features · Penn Action | accuracy (%): 98.1 -> 98.2 | Improved |
| add decoupled action poses (ablation) | Combined features · MPII val | PCKh (%): 85.1 -> 85.4 | Improved |
| add joint confidence scores to weight poses (ablation) | Pose features · Penn Action | accuracy (%) | No clear effect |
Closest papers by meaning
- 2D/3D Pose Estimation and Action Recognition Using Multitask Deep Learning (CVPR 2018 · similarity 0.95)
- Deep Multitask Architecture for Integrated 2D and 3D Human Sensing (CVPR 2017 · similarity 0.90)
- Lifting From the Deep: Convolutional 3D Pose Estimation From a Single Image (CVPR 2017 · similarity 0.89)
- A Simple yet Effective Baseline for 3D Human Pose Estimation (ICCV 2017 · similarity 0.89)
- PA3D: Pose-Action 3D Machine for Video Recognition (CVPR 2019 · similarity 0.89)
- Coarse-To-Fine Volumetric Prediction for Single-Image 3D Human Pose (CVPR 2017 · 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.
