Research map / 3D vision / Human pose estimation
Attention Mechanism Exploits Temporal Contexts: Real-Time 3D Human Pose Reconstruction
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the attention-based temporal convolution model | Video-based 3D pose estimation with CPN 2D input · Human3.6M Protocol#1 | MPJPE (mm): 46.8 -> 45.1 | Helped |
| Use ground-truth rather than detected 2D joints as input | Ours (n=243) · Human3.6M Protocol#1 | MPJPE (mm): 45.1 -> 34.7 | Helped |
| Add the attention-based temporal convolution model | Video-based 3D pose estimation with CPN 2D input · Human3.6M Protocol#2 | P-MPJPE (mm): 36.5 -> 35.6 | Helped |
| Add the proposed model | Video-based 3D pose estimation · HumanEva Protocol#2 | P-MPJPE (mm): 15.8 -> 15.4 | Helped |
| Combine attention and multi-scale dilation rather than using either alone (ablation) | Proposed pose model · Human3.6M Protocol#1, CPN 2D input, n=243 | MPJPE (mm): 46.3 -> 45.1; 45.7 -> 45.1 | Helped |
| Combine attention and multi-scale dilation rather than using either alone (ablation) | Proposed pose model · Human3.6M Protocol#1, ground-truth 2D input, n=243 | MPJPE (mm): 35.5 -> 34.7; 35.3 -> 34.7 | Helped |
| Use kernel attention with M=3 instead of M=2 or M=4 (ablation) | Kernel attention module · Human3.6M Protocol#1 | MPJPE (mm): 37.1 -> 35.5; 38.0 -> 35.5 | Helped |
| Set kernel-attention groups to G=2 rather than G=8 (ablation) | Kernel attention module · Human3.6M Protocol#1 | MPJPE (mm): 35.5 -> 35.3 | Helped |
| Set kernel-attention groups to G=1 or G=16 rather than G=8 (ablation) | Kernel attention module · Human3.6M Protocol#1 | MPJPE (mm): 35.5 -> 37.4; 35.5 -> 37.3 | Hurt |
| Vary kernel-attention reduction dimension from r=128 (ablation) | Kernel attention module · Human3.6M Protocol#1 | MPJPE (mm): 35.5 -> 35.9; 35.5 -> 36.2 | Hurt |
| Increase receptive field from n=27 to n=147 in the tested shallow configuration (ablation) | Proposed pose model · Human3.6M Protocol#1 | MPJPE (mm): 40.6 -> 36.8 | Helped |
| Increase receptive field to n=1029 instead of using n=243 (ablation) | Proposed pose model · Human3.6M Protocol#1 | MPJPE (mm): 35.5 -> 37.0 | Hurt |
Closest papers by meaning
- A Single 2D Pose with Context is Worth Hundreds for 3D Human Pose Estimation (NeurIPS 2023 · similarity 0.90)
- Capturing Humans in Motion: Temporal-Attentive 3D Human Pose and Shape Estimation From Monocular Video (CVPR 2022 · similarity 0.90)
- Deep Reinforcement Learning for Active Human Pose Estimation (AAAI 2020 · similarity 0.90)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.90)
- Deep Dual Consecutive Network for Human Pose Estimation (CVPR 2021 · similarity 0.90)
- PoseMamba: Monocular 3D Human Pose Estimation with Bidirectional Global-Local Spatio-Temporal State Space Model (AAAI 2025 · 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.