Research map / 3D vision / Human pose estimation
ViPNAS: Efficient Video Pose Estimation via Neural Architecture Search
Lumin Xu, Yingda Guan, Sheng Jin, Wentao Liu, Chen Qian, Ping Luo, Wanli Ouyang, Xiaogang Wang
Check 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 |
|---|---|---|---|
| Search temporal-level architectures and feature fusion under the same computation budget (ablation) | ViPNAS · PoseTrack2018 val, ResNet-50, 0.82 GFLOPs | mAP: 80.3 -> 81.6 | Improved |
| Search temporal-level architectures and feature fusion under the same computation budget (ablation) | ViPNAS · PoseTrack2018 val, MobileNet-V3, 0.37 GFLOPs | mAP: 77.2 -> 78.2 | Improved |
| Add temporal feature fusion (ablation) | T-ViPNAS · PoseTrack2018 val | mAP | Improved |
| Allow different frame-specialized architectures instead of sharing the same spatial and temporal architectures (ablation) | T-ViPNAS · PoseTrack2018 val | mAP: at least 0.5% mAP | Improved |
| Vary propagation length from 2 to 4 frames (ablation) | ViPNAS temporal search · PoseTrack2018 val, 1.0 GFLOPs | mAP: 81.7; 81.7; 81.4 | No clear effect |
| Replace SBL with S-ViPNAS | MobileNet-V3 pose estimator · PoseTrack2018 val | Total AP: 78.1 -> 81.1 | Improved |
| Replace SBL with T-ViPNAS | MobileNet-V3 pose estimator · PoseTrack2018 val | GFLOPs: 4.1 -> 0.37 | Improved |
| Search a compact ResNet-50 architecture | S-ViPNAS · COCO2017 val2017 | AP: 70.4 -> 71.0 | Improved |
| Search a compact MobileNet-V3 architecture | S-ViPNAS · COCO2017 val2017 | AP: 64.7 -> 67.8 | Improved |
| Search the HRNet-W32 architecture | S-ViPNAS · COCO2017 val2017 | AP: 74.4 -> 74.7 | Improved |
| Search the HRNet-W32 architecture | S-ViPNAS · COCO2017 val2017 | GFLOPs: 7.10 -> 5.64 | Improved |
Closest papers by meaning
- Enhanced 3D Human Pose Estimation from Videos by Using Attention-Based Neural Network with Dilated Convolutions (IJCV 2021 · similarity 0.87)
- Attention Mechanism Exploits Temporal Contexts: Real-Time 3D Human Pose Reconstruction (CVPR 2020 · similarity 0.87)
- Exploiting temporal information for 3D human pose estimation (ECCV 2018 · similarity 0.87)
- Learning Temporal Pose Estimation from Sparsely-Labeled Videos (NeurIPS 2019 · similarity 0.87)
- Neural Architecture Search for Joint Human Parsing and Pose Estimation (ICCV 2021 · similarity 0.87)
- ZoomNAS: Searching for Whole-body Human Pose Estimation in the Wild (TPAMI 2022 · similarity 0.86)
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.
