Research map / 3D vision / Human pose estimation
Enhanced 3D Human Pose Estimation from Videos by Using Attention-Based Neural Network with Dilated Convolutions
Ruixu Liu, Ju Shen, He Wang, Chen Chen, Sen-ching S. Cheung, Vijayan K. Asari
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 temporal and kernel attention to a plain 243-frame temporal convolution (ablation) | TCN · Human3.6M (GT 2D) | MPJPE (mm): 37.8 -> 35.5 | Improved |
| add multi-scale dilation on top of attention (ablation) | ATCN · Human3.6M (GT 2D) | MPJPE (mm): 35.5 -> 34.7 | Improved |
| add multi-scale dilation to attention model (n=243) (ablation) | ATCN · Human3.6M (GT 2D, n=243) | MPJPE (mm): 35.5 -> 33.4 | Improved |
| add attention to multi-scale dilation model (n=243) (ablation) | MDC · Human3.6M (CPN 2D, n=243) | MPJPE (mm): 49.1 -> 45.1 | Improved |
| enlarge receptive field from 27 to 243 frames (ablation) | baseline TCN · Human3.6M (CPN 2D) | MPJPE (mm): 51.2 -> 49.2 | Improved |
| add data augmentation with virtual cameras (ablation) | ATCN+MDC · Human3.6M (CPN 2D) | MPJPE (mm): 46.5 -> 44.8 | Improved |
| increase kernel attention branches M from 3 to 4 (ablation) | kernel attention · Human3.6M (GT 2D, n=243) | MPJPE (mm): 35.5 -> 38.0 | Hurt |
| use M=3 branches instead of M=1 (ablation) | kernel attention · Human3.6M (GT 2D, n=243) | MPJPE (mm): 37.8 -> 35.5 | Improved |
| use 16 groups instead of 8 (ablation) | kernel attention · Human3.6M (GT 2D, n=243) | MPJPE (mm): 35.5 -> 37.3 | Hurt |
| reduce groups from 8 to 2 (ablation) | kernel attention · Human3.6M (GT 2D, n=243) | MPJPE (mm): 35.5 -> 35.3 | No clear effect |
| increase receptive field to 1029 frames (ablation) | ATCN · Human3.6M (GT 2D) | MPJPE (mm): 35.5 -> 37.0 | Hurt |
| increase receptive field from 27 to 147 frames (3 layers) (ablation) | ATCN · Human3.6M (GT 2D) | MPJPE (mm): 40.6 -> 36.8 | Improved |
Closest papers by meaning
- Attention Mechanism Exploits Temporal Contexts: Real-Time 3D Human Pose Reconstruction (CVPR 2020 · similarity 0.94)
- Exploiting temporal information for 3D human pose estimation (ECCV 2018 · similarity 0.90)
- A Single 2D Pose with Context is Worth Hundreds for 3D Human Pose Estimation (NeurIPS 2023 · similarity 0.89)
- A Simple yet Effective Baseline for 3D Human Pose Estimation (ICCV 2017 · similarity 0.89)
- From Human Pose Similarity Metric to 3D Human Pose Estimator: Temporal Propagating LSTM Networks (TPAMI 2022 · similarity 0.88)
- Multi-Context Attention for Human Pose Estimation (CVPR 2017 · similarity 0.88)
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.
