Downloads · 30 days
0
VuNgocDuong/depose-fi-onnx
depose-fi-onnx is a machine learning model from VuNgocDuong. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
Decomposition-first WiFi human pose estimation. Each CSI frame's amplitude and phase are separately CP-factorized (phase sanitized via cross-antenna phase linking + linear-trend removal), then fused by Selective Compo…
Downloads · 30 days
0
Access
Public
Updated Jul 26, 2026
Repo size
15.6 MB
Likes
0
Public
Click a slice to open those files.
.data7.6 MB · 93%
From the Hugging Face model README
Decomposition-first WiFi human pose estimation. Each CSI frame's amplitude and phase are separately CP-factorized (phase sanitized via cross-antenna phase linking + linear-trend removal), then fused by Selective Component-Adaptive Fusion (S-AFF). See the paper/code: https://github.com/Duongvu05/DePose-Fi
Both models take two inputs (cp_factors_amp, cp_factors_phase),
each the canonicalized + zero-padded CP factors at the model's R_max,
produced by deploy/run_inference.py in the repo above (do not feed raw
CSI directly -- these models start after CP decomposition).
| File | Dataset | R_max | Params | Headline N-MPJPE |
|---|---|---|---|---|
saff_dualcp_rmax.onnx (+ .onnx.data) | MM-Fi (single-person) | 20 | 190.9K | 36.35% (at R=10) |
saff_dualcp_piw3d_rmax.onnx (+ .onnx.data) | Person-in-WiFi 3D (single-person) | 24 | 1.68M | 21.35% (at R=16) |
N-MPJPE is torso-normalized (%), not raw millimeters.
See deploy/README.md in the GitHub repo for the full edge-deployment
pipeline (Raspberry Pi / Jetson Nano / Qualcomm QCS6490, PRA rank/iteration
adaptation). Quick local test:
python deploy/run_inference.py --device raspberry_pi \
--model saff_dualcp_rmax.onnx \
--deadline-ms 30 \
--frames "path/to/wifi-csi/*.mat"