echopose-sdk

WiFi CSI pose-estimation toolkit — the Python companion to the EchoPose system.

EchoPose turns commodity ESP32-S3 access points into a privacy-preserving human pose estimator that requires no cameras, no wearables, and works through walls. This SDK provides the tools you need to collect, process, evaluate, and replay CSI data and pose predictions.

Features

Installation

pip install echopose-sdk

Development install (editable):

pip install -e echopose_sdk/

Quick start

Validate a bundle

from echopose_sdk import validate_bundle
ok, reason = validate_bundle(bundle_dict)

Compute pose metrics

import numpy as np
from echopose_sdk.metrics import summary_report

pred = np.load("pred_poses.npy")   # shape (N, 17, 3)
gt   = np.load("gt_poses.npy")

report = summary_report(pred, gt)
# {'mpjpe': 0.42, 'pck_01_abs': 0.31, 'body_pck_01': 0.68, 'pa_mpjpe': 0.29, ...}

Normalise CSI amplitudes

from echopose_sdk.csi import normalize_subcarriers
norm = normalize_subcarriers(raw_amplitudes, method="zscore")

Smooth a skeleton sequence

from echopose_sdk.skeleton import smooth_skeleton_sequence
smoothed = smooth_skeleton_sequence(seq, method="gaussian", window=7)

Stream a session file

from echopose_sdk.streaming import BundleReader

with BundleReader("session.jsonl") as reader:
    for bundle in reader:
        print(bundle["timestamp_ms"])

CLI

echopose-sdk validate  <bundle.json>
echopose-sdk inspect   <bundle.json> [--node NODE] [--pretty]
echopose-sdk metrics   --pred pred.npy --gt gt.npy [--per-joint] [--pretty]
echopose-sdk stream    <session.jsonl> [--fps 20] [--limit 100]

License

MIT © Muhammed Shazin Sadhik Kunhi Parambath