Research map / 3D vision / Human pose estimation
HPE-Li: WiFi-enabled Lightweight Dual Selective Kernel Convolution for Human Pose Estimation
Toan D. Gian, Tien Dac Lai, Thien Van Luong, Kok-Seng Wong, Van-Dinh Nguyen
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 frequency-wise selective kernel attention on top of channel-wise SK convolution (ablation) | CNN with SKConv · MM-Fi P3-S2 | MPJPE (mm): 210.45 -> 182.62 | Improved |
| replace standard convolution with DSKConv (ablation) | CNN backbone · MM-Fi P3-S2 | MPJPE (mm): 236.59 -> 182.62 | Improved |
| use DSKConv instead of dynamic convolution (ablation) | CNN backbone · MM-Fi P3-S2 | MPJPE (mm): 218.92 -> 182.62 | Improved |
| use multiple kernel branches (N=4) instead of one (ablation) | DSKNet · MM-Fi P3-S1 | MPJPE (mm): 152.67 -> 149.43 | Improved |
| increase branches from 4 to 5 (ablation) | DSKNet · MM-Fi P3-S1 | MPJPE (mm): 149.43 -> 150.21 | Hurt |
| increase branches from 3 to 4 (ablation) | DSKNet · MM-Fi P3-S1 | MPJPE (mm): 149.48 -> 149.43 | No clear effect |
| use dilated 3x3 (D=2) instead of undilated 5x5 kernel (ablation) | DSKConv · MM-Fi P3-S1 | MPJPE (mm): 149.71 -> 149.43 | Improved |
| increase group number G from 32 to 128 (ablation) | DSKConv · MM-Fi P3-S1 | MPJPE (mm): 150.57 -> 150.97 | No clear effect |
| set reduction ratio r to 32 versus 1 (ablation) | DSKNet · MM-Fi P3-S1 | MPJPE (mm): 151.85 -> 149.43 | Improved |
| add uniform or median denoising filter under AWGN and salt-and-pepper noise (ablation) | HPE-Li · MM-Fi P3-S1 with noise | MPJPE (mm) | Improved |
| use plain MSE loss instead of pose adjacency matrix loss | HPE-Li · MM-Fi and WiPose | pose accuracy | Improved |
Later papers that compared with this method
2 later accepted papers report HPE-Li as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Interleaved Selective State Space Models for Efficient WiFi-Based 3D Multi-Person Pose Estimation (ICML 2026 · better on 4 of 4 shared results)
- WiFi-JEPA: Self-supervised Learning for WiFi-CSI 3D Human Pose Estimation (ECCV 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- WiFi-JEPA: Self-supervised Learning for WiFi-CSI 3D Human Pose Estimation (ECCV 2026 · similarity 0.87)
- Interleaved Selective State Space Models for Efficient WiFi-Based 3D Multi-Person Pose Estimation (ICML 2026 · similarity 0.86)
- Through-Wall Human Pose Estimation Using Radio Signals (CVPR 2018 · similarity 0.85)
- Lite Pose: Efficient Architecture Design for 2D Human Pose Estimation (CVPR 2022 · similarity 0.85)
- Deep Semantic Graph Transformer for Multi-View 3D Human Pose Estimation (AAAI 2024 · similarity 0.83)
- Person-in-WiFi 3D: End-to-End Multi-Person 3D Pose Estimation with Wi-Fi (CVPR 2024 · similarity 0.83)
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.
