Research map / 3D vision / Human pose estimation
WiFi-JEPA: Self-supervised Learning for WiFi-CSI 3D Human Pose Estimation
Doeon Kim, Jungyoon Lee, SEONGSIN KIM, Seong-heum Kim
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 |
|---|---|---|---|
| use link masking instead of random masking (ablation) | WiFi-JEPA · PiW3D (real 90K pre-training) | MPJPE (mm): 105.14 -> 97.10 | Improved |
| use link masking instead of multi-block masking (ablation) | WiFi-JEPA · PiW3D (real 90K pre-training) | MPJPE (mm): 102.06 -> 97.10 | Improved |
| use link masking instead of time masking (ablation) | WiFi-JEPA · PiW3D (real 90K pre-training) | MPJPE (mm): 104.24 -> 97.10 | Improved |
| add CSI-specific (C,T,L) tokenization instead of 2D spectrogram flattening (ablation) | WiFi-JEPA · PiW3D (real 90K pre-training) | MPJPE (mm): 111.85 -> 97.1 | Improved |
| add CSI-specific (C,T,L) tokenization instead of 2D spectrogram flattening (ablation) | WiFi-JEPA · PiW3D (real 90K pre-training) | PA-MPJPE (mm): 70.98 -> 67.20 | Improved |
| add JEPA pre-training on real CSI versus training from scratch (ablation) | WiFi-JEPA · PiW3D | MPJPE (mm): 102.4 -> 97.1 | Improved |
| pre-train with MAE instead of no pre-training | ViT + PETR decoder · PiW3D | MPJPE (mm): 102.4 -> 130.3 | Hurt |
| pre-train with SimMIM, BYOL or MoCoV3 instead of no pre-training | ViT + PETR decoder · PiW3D | MPJPE (mm): 102.4 -> 145.6 | Hurt |
| add sim-object (geometric primitive) simulated CSI pre-training only (ablation) | WiFi-JEPA · PiW3D | MPJPE (mm): 102.4 -> 100.1 | Improved |
| pre-train on sim-human (human-mesh) simulated CSI instead of sim-object (ablation) | WiFi-JEPA · PiW3D | MPJPE (mm): 100.1 -> 110.3 | Hurt |
| add 90K sim-object frames to 90K real frames for pre-training (ablation) | WiFi-JEPA · PiW3D | MPJPE (mm): 97.1 -> 93.5 | Improved |
| halve real+sim pre-training data to 45K each (ablation) | WiFi-JEPA · PiW3D | MPJPE (mm): 93.5 -> 97.2 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- HPE-Li — HPE-Li: WiFi-enabled Lightweight Dual Selective Kernel Convolution for Human Pose Estimation (ECCV 2024) · 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
- Interleaved Selective State Space Models for Efficient WiFi-Based 3D Multi-Person Pose Estimation (ICML 2026 · similarity 0.91)
- Person-in-WiFi 3D: Unified Model for 3D WiFi Perception (TPAMI 2026 · similarity 0.88)
- Person-in-WiFi 3D: End-to-End Multi-Person 3D Pose Estimation with Wi-Fi (CVPR 2024 · similarity 0.86)
- CSI-Bench: A Large-Scale In-the-Wild Dataset for Multi-task WiFi Sensing (NeurIPS 2025 · similarity 0.84)
- Hourglass Tokenizer for Efficient Transformer-Based 3D Human Pose Estimation (CVPR 2024 · similarity 0.83)
- ConSense: Continually Sensing Human Activity with WiFi via Growing and Picking (AAAI 2025 · 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.
