CSI → pose (skeleton) overlaid on your laptop camera
+
+
+
+
+
+
+
+
+
camera: off
+
Camera is a visual reference only — it is NOT fed to the model. Overlay alignment is approximate (model trained in a different camera's frame).
+
+
+
live
+
CSI source—
+
nodes—
+
presence—
+
motion—
+
pose fps—
+
+ This skeleton is inferred from WiFi CSI only — no camera in the loop here. A model was
+ trained on paired (camera-pose, CSI) data in this room (ADR-079/180).
+
+ Honest accuracy: ~59.5% PCK@0.10 on held-out data (vs a 50% mean-pose baseline →
+ +9.4 pp real signal). It captures coarse pose; fine detail is weak (PCK@0.05 ≈ 24%).
+ Same person / room / session — not validated cross-day or through-wall.
+
+
+
+
+
+
diff --git a/examples/through-wall/wiflow_ab.py b/examples/through-wall/wiflow_ab.py
new file mode 100644
index 00000000..f452ca35
--- /dev/null
+++ b/examples/through-wall/wiflow_ab.py
@@ -0,0 +1,126 @@
+#!/usr/bin/env python3
+"""Rigorous A/B for WiFlow CSI->pose: is the held-out PCK real signal or split leakage?
+
+For a dataset of {csi:[D], kps:17x[x,y,vis]} pairs, train the SAME small MLP under
+several train/val SPLITS and report held-out PCK@0.10 vs the mean-pose baseline:
+
+ - chronological_80_20 : last 20% in time (val temporally ADJACENT to train -> leaks
+ via CSI/pose autocorrelation; this is what gave us +9.4)
+ - random_80_20 : shuffled (val frames interleaved with train -> MAX leak)
+ - blocked_gap : hold out a contiguous MIDDLE block with a time GAP buffer on
+ each side so val is NOT adjacent to any train frame -> the
+ honest, leakage-controlled test
+
+If the model beats baseline on chronological/random but COLLAPSES to ~baseline on
+blocked_gap, the apparent signal was temporal leakage, not generalizable CSI->pose.
+
+Usage (ruvultra venv): python wiflow_ab.py --data ~/wiflow-room/dataset.jsonl
+"""
+import argparse, json, sys
+import numpy as np, torch, torch.nn as nn
+
+def _rec(r, X, Y, V, B):
+ X.append(r["csi"]); kp=r["kps"]
+ if kp and isinstance(kp[0], (list,tuple)): # 17 x [x,y(,vis)]
+ Y.append([c for k in kp for c in (k[0],k[1])]); V.append([(k[2] if len(k)>2 else 1.0) for k in kp])
+ else: # flat 34 (browser export, no vis)
+ Y.append(list(kp)); V.append([1.0]*17)
+ B.append(r.get("bucket"))
+
+def load(path):
+ X,Y,V,B=[],[],[],[]
+ txt=open(path).read().strip()
+ if txt[:1] in "[{": # JSON (browser export: dict{samples:[]} or bare array)
+ d=json.loads(txt)
+ rows = d if isinstance(d,list) else d.get("samples", d.get("data", []))
+ for r in rows: _rec(r,X,Y,V,B)
+ else: # JSONL (python capture)
+ for line in txt.splitlines():
+ if line.strip(): _rec(json.loads(line),X,Y,V,B)
+ return np.array(X,np.float32), np.array(Y,np.float32), np.array(V,np.float32), B
+
+class Net(nn.Module):
+ def __init__(s,din,dout):
+ super().__init__()
+ s.n=nn.Sequential(nn.Linear(din,384),nn.ReLU(),nn.Dropout(.35),
+ nn.Linear(384,192),nn.ReLU(),nn.Dropout(.35),
+ nn.Linear(192,96),nn.ReLU(),nn.Linear(96,dout),nn.Sigmoid())
+ def forward(s,x): return s.n(x)
+
+def pck(pred,gt,vis,thr=0.10):
+ p=pred.reshape(-1,17,2); g=gt.reshape(-1,17,2)
+ d=np.linalg.norm(p-g,axis=2); m=vis>0.5
+ return float((d[m] val poses/activities never seen in train.
+ # the strictest leakage-free test (only when bucket labels exist).
+ b=np.array([x if x is not None else -1 for x in B])
+ uniq=[u for u in sorted(set(b.tolist())) if u!=-1]
+ if len(uniq)<3: raise ValueError("too few buckets")
+ hold=set(uniq[::max(1,len(uniq)//3)][:max(1,len(uniq)//3)]) # ~1/3 of activities held out
+ val=idx[np.isin(b,list(hold))]; train=idx[~np.isin(b,list(hold))]
+ return train, val
+ raise ValueError(kind)
+
+def run(X,Y,V,tr,va,epochs=250,seed=0):
+ torch.manual_seed(seed); np.random.seed(seed) # seed weight init + batch shuffle
+ dev="cuda" if torch.cuda.is_available() else "cpu"
+ mu,sd=X[tr].mean(0),X[tr].std(0)+1e-6
+ Xtr=torch.tensor((X[tr]-mu)/sd).to(dev); Ytr=torch.tensor(Y[tr]).to(dev)
+ Xva=torch.tensor((X[va]-mu)/sd).to(dev)
+ net=Net(X.shape[1],Y.shape[1]).to(dev)
+ opt=torch.optim.Adam(net.parameters(),lr=1e-3,weight_decay=1e-4); lf=nn.MSELoss()
+ best=(1e9,None)
+ for ep in range(epochs):
+ net.train(); perm=torch.randperm(len(Xtr),device=dev)
+ for i in range(0,len(Xtr),64):
+ j=perm[i:i+64]; opt.zero_grad(); loss=lf(net(Xtr[j]),Ytr[j]); loss.backward(); opt.step()
+ net.eval()
+ with torch.no_grad(): pv=net(Xva).cpu().numpy()
+ vl=float(((pv-Y[va])**2).mean())
+ if vl16}{'baseline':>11}{'delta (mean±sd)':>20} verdict")
+ print("-"*86)
+ splits=["chronological_80_20","random_80_20","blocked_gap"]+(["grouped_bucket"] if has_buckets else [])
+ for kind in splits:
+ try:
+ tr,va=split_idx(n,kind,B)
+ ms=[]; bs=[]
+ for s in range(a.seeds):
+ m,b=run(X,Y,V,tr,va,a.epochs,seed=s); ms.append(m); bs.append(b)
+ ms=np.array(ms)*100; bs=np.array(bs)*100; ds=ms-bs
+ dm,dsd=ds.mean(),ds.std()
+ # REAL only if the mean delta minus 1 sd still clears the 1.5pp threshold (robust to seed variance)
+ verdict = "REAL signal" if dm-dsd>1.5 else ("weak/uncertain" if dm>1.5 else "no signal (==baseline)")
+ print(f"{kind:<22}{ms.mean():>13.1f}±{ms.std():>3.1f}{bs.mean():>10.1f}%{dm:>+12.1f}±{dsd:>4.1f}pp {verdict}")
+ except Exception as e:
+ print(f"{kind:<22} skipped: {e}")
+ print(f"\nmean±sd over {a.seeds} seeds (weight init + batch order). blocked_gap = 10% time gap each")
+ print("side; grouped_bucket holds out ENTIRE activities (strictest). If only the LEAKY splits")
+ print("(chronological/random) beat baseline, the apparent signal is leakage, not generalizable pose.")
+
+if __name__=="__main__": main()
diff --git a/examples/through-wall/wiflow_browser.html b/examples/through-wall/wiflow_browser.html
index 72e88119..5d02aaaf 100644
--- a/examples/through-wall/wiflow_browser.html
+++ b/examples/through-wall/wiflow_browser.html
@@ -112,7 +112,11 @@
empty-room baseline (ADR-151) — step OUT of the space
-
+
+ not detected
+
+
+
@@ -285,9 +289,15 @@
// wss when served over https (mobile/secure-context safe), else ws; ?ws= overrides
const CSI_WS = (new URLSearchParams(location.search)).get('ws')
|| `${location.protocol === 'https:' ? 'wss' : 'ws'}://${location.hostname || 'localhost'}:8765/ws/sensing`;
-const NODE_IDS = [9, 13]; // per-node features in this fixed order (matches Python pipeline)
+// Per-node feature schema — AUTO-DETECTED from the live stream (see detectSensors).
+// [9,13] is only the fallback until detection runs. ORDER is fixed (sorted ascending)
+// so the model's input layout is stable across capture / train / infer.
+let NODE_IDS = [9, 13];
const FIELD_LEN = 400; // signal_field.values padded/truncated to 400
-const CSI_DIM = 4 + NODE_IDS.length * 3 + FIELD_LEN; // 4 + 6 + 400 = 410
+let CSI_DIM = 4 + NODE_IDS.length * 3 + FIELD_LEN; // 4 global + 3/node + 400 field
+function recomputeCsiDim(){ CSI_DIM = 4 + NODE_IDS.length * 3 + FIELD_LEN; }
+let sensorsDetected = false; // true once a detect (auto/manual/restored) has locked the node set
+let autoDetectStarted = false; // one-shot guard for the auto-detect on first live frame
const N_KP = 17, OUT_DIM = N_KP * 2; // 17 COCO keypoints -> 34 coords
const BASELINE_SECONDS = 10; // empty-room calibration window
const EPS = 1e-6;
@@ -333,9 +343,9 @@ async function selectBackend(){
// ============================================================================
// CSI vector construction — MUST match wiflow_capture.py csi_vector() exactly.
// [mean_rssi, variance, motion_band_power, breathing_band_power] (4 global)
-// + for node 9 then node 13: [mean_rssi, variance, motion_band_power] (6 per-node)
+// + for each node in NODE_IDS order: [mean_rssi, variance, motion_band_power] (3 per-node)
// + signal_field.values padded/truncated to 400 (400 field)
-// = 410-d (RAW — baseline-normalization applied separately, see baselineNorm)
+// = CSI_DIM-d (RAW — baseline-normalization applied separately, see baselineNorm)
// ============================================================================
function csiVector(frame){
const f = frame.features || {};
@@ -368,6 +378,87 @@ function baselineNorm(vecRaw){
return out;
}
+// ============================================================================
+// ESP32 sensor auto-detection
+// Sniff the live /ws/sensing stream, find which node_ids are actually present
+// and healthy, and lock that ordered set as the per-node schema (NODE_IDS/CSI_DIM).
+// The node set defines the model's input dimension, so detection must run BEFORE
+// calibration + capture; changing it invalidates a baseline/dataset built on a
+// different set (we confirm, then reset, on a manual re-detect).
+// ============================================================================
+async function detectSensors(ms = 3000){
+ const tally = {}; // node_id -> { seen, fps, rssi }
+ let frames = 0;
+ const t0 = performance.now();
+ const el = $('detNodes'); if (el){ el.textContent = 'scanning…'; el.className = 'v'; }
+ while (performance.now() - t0 < ms){
+ if (latestCSI.frame && latestCSI.source === 'esp32'){
+ frames++;
+ for (const nf of (latestCSI.frame.node_features || [])){
+ const id = nf.node_id; if (id == null) continue;
+ const f = nf.features || {};
+ const t = (tally[id] || (tally[id] = { seen:0, fps:0, rssi:0 }));
+ t.seen++; t.fps += (+nf.frame_rate_hz || 0);
+ t.rssi += (+f.mean_rssi || +nf.rssi_dbm || 0);
+ }
+ }
+ await new Promise(r => setTimeout(r, 100));
+ }
+ // healthy = seen in >40% of sampled frames (filters transient / duplicate ids)
+ const healthy = Object.keys(tally).map(k => ({
+ id:+k, seen:tally[k].seen, fps:tally[k].fps/tally[k].seen, rssi:tally[k].rssi/tally[k].seen }))
+ .filter(n => n.seen >= Math.max(2, frames * 0.4))
+ .sort((a,b)=> a.id - b.id);
+ return { healthy, frames };
+}
+
+function renderDetectedSensors(list){
+ const el = $('detNodes'); if (!el) return;
+ el.textContent = list.length
+ ? list.map(n => `#${n.id} (${Math.round(n.fps)}fps, ${Math.round(n.rssi)}dB)`).join(' · ')
+ : 'none found';
+ el.className = list.length ? 'v green' : 'v red';
+}
+
+async function runDetect(manual){
+ const { healthy, frames } = await detectSensors(manual ? 4000 : 3000);
+ if (!healthy.length){
+ const el = $('detNodes');
+ if (el){ el.textContent = frames ? 'no healthy nodes' : 'no live CSI (start sensing-server / esp32)';
+ el.className = 'v red'; }
+ return;
+ }
+ const ids = healthy.map(n => n.id);
+ const changed = ids.length !== NODE_IDS.length || ids.some((v,i)=> v !== NODE_IDS[i]);
+ if (changed && (baseline || SAMPLES.length)){
+ const ok = confirm(
+ `Detected sensors [${ids.join(', ')}] differ from the current set [${NODE_IDS.join(', ')}].\n\n` +
+ `The node set defines the model input, so switching invalidates the existing baseline` +
+ (SAMPLES.length ? ` and ${SAMPLES.length} captured samples` : ``) +
+ `. Reset and use the detected set?`);
+ if (!ok){ renderDetectedSensors(healthy); return; }
+ if (baseline){ baseline = null; stageDone.calibrate = false; idbDel('baseline');
+ $('calStatus').textContent = 'NOT CALIBRATED'; $('calStatus').className = 'v'; $('calBar').style.width = '0%'; }
+ if (SAMPLES.length){ SAMPLES = []; covCounts = new Array(BUCKETS.length).fill(0);
+ idbPut('samples', []); $('capN').textContent = '0'; $('trN').textContent = '0'; renderCoverage(); }
+ }
+ NODE_IDS = ids; recomputeCsiDim(); sensorsDetected = true;
+ idbPut('nodeIds', NODE_IDS);
+ renderDetectedSensors(healthy);
+ refreshGates();
+}
+
+async function restoreNodeIds(){
+ try{
+ const ids = await idbGet('nodeIds');
+ if (Array.isArray(ids) && ids.length){
+ NODE_IDS = ids.slice(); recomputeCsiDim(); sensorsDetected = true;
+ const el = $('detNodes');
+ if (el){ el.textContent = 'restored: ' + NODE_IDS.map(i => '#' + i).join(' '); el.className = 'v'; }
+ }
+ }catch(e){ /* ignore */ }
+}
+
// ============================================================================
// CSI WebSocket
// ============================================================================
@@ -388,6 +479,11 @@ function connectCSI(){
source: src,
nodes: (d.nodes || []).map(n => n.node_id).filter(x => x != null).sort((a,b)=>a-b)
};
+ // auto-detect the sensor set once, on the first live frame, only when starting fresh
+ // (no baseline / no samples) so we never silently change a schema work is built on.
+ if (src === 'esp32' && !sensorsDetected && !autoDetectStarted && !baseline && SAMPLES.length === 0){
+ autoDetectStarted = true; runDetect(false);
+ }
if (src === 'esp32') banner('live','LIVE — real ESP32 CSI');
else banner('sim',`SIMULATED — not real (source=${src})`);
};
@@ -599,6 +695,7 @@ function finishCalibration(){
refreshGates();
}
$('calBtn').addEventListener('click', startCalibration);
+$('detBtn').addEventListener('click', ()=> runDetect(true));
$('recalBtn').addEventListener('click', ()=>{ baseline = null; stageDone.calibrate = false;
$('calStatus').textContent = 'NOT CALIBRATED'; $('calStatus').className = 'v';
$('calBar').style.width = '0%'; $('calN').textContent = '0'; idbDel('baseline'); refreshGates(); startCalibration(); });
@@ -736,7 +833,7 @@ $('clrBtn').addEventListener('click', async ()=>{
$('expBtn').addEventListener('click', ()=>{
const out = {
format: 'wiflow-browser-dataset', version: 1, exported: new Date().toISOString(),
- csi_dim: CSI_DIM, out_dim: OUT_DIM, buckets: BUCKETS,
+ csi_dim: CSI_DIM, out_dim: OUT_DIM, buckets: BUCKETS, nodes: NODE_IDS.slice(),
note: 'csi is baseline-normalized (ADR-151 deviation-from-baseline); kps are 17 COCO keypoints in [0,1] image coords',
samples: SAMPLES.map((s,i)=>({ csi: Array.from(s.csi), kps: Array.from(s.kps), bucket: s.bucket, t: (s.t!=null?s.t:i) }))
};
@@ -1152,6 +1249,7 @@ function inferLoop(){
(async function boot(){
connectCSI();
await selectBackend();
+ await restoreNodeIds(); // restore a previously-detected sensor set (fixes CSI_DIM before baseline)
await loadBaseline();
await idbLoad();
await loadModel();
diff --git a/examples/through-wall/wiflow_capture.py b/examples/through-wall/wiflow_capture.py
new file mode 100644
index 00000000..b77c3951
--- /dev/null
+++ b/examples/through-wall/wiflow_capture.py
@@ -0,0 +1,161 @@
+#!/usr/bin/env python3
+"""WiFlow-style camera-supervised capture (ADR-079 / ADR-180).
+
+Runs on a box with BOTH a camera (ground truth) and reachable live CSI:
+ - opens a camera, runs MediaPipe Pose -> 17 COCO keypoints (the LABEL),
+ - subscribes to the sensing-server /ws/sensing (the INPUT: CSI features +
+ 20x20 signal-field),
+ - writes timestamp-aligned (csi -> pose) pairs to a JSONL dataset.
+
+This is the *collect* phase of camera-supervised CSI->pose training. The camera
+and the CSI nodes MUST see the same person in the same space at the same time,
+or the pairs are meaningless. Honest by construction: we only emit a pair when
+BOTH a confident camera pose AND a live (source=esp32) CSI frame are present in
+the same ~100 ms window.
+
+Usage (on ruvultra, with the CSI tunneled to localhost:8765):
+ python3 wiflow_capture.py --ws ws://localhost:8765/ws/sensing \
+ --cam 0 --out ~/wiflow-room/dataset.jsonl --seconds 180
+"""
+import argparse, asyncio, json, time, threading, sys, os
+from collections import deque
+
+import urllib.request
+import cv2
+import numpy as np
+import mediapipe as mp
+from mediapipe.tasks.python import BaseOptions
+from mediapipe.tasks.python.vision import PoseLandmarker, PoseLandmarkerOptions, RunningMode
+import websockets
+
+_MODEL_URL = ("https://storage.googleapis.com/mediapipe-models/pose_landmarker/"
+ "pose_landmarker_lite/float16/latest/pose_landmarker_lite.task")
+
+def ensure_model(path: str) -> str:
+ if not os.path.exists(path):
+ os.makedirs(os.path.dirname(path), exist_ok=True)
+ print(f"[capture] downloading pose model -> {path}", flush=True)
+ urllib.request.urlretrieve(_MODEL_URL, path)
+ return path
+
+# MediaPipe Pose (33 landmarks) -> 17 COCO keypoints (same mapping as
+# scripts/collect-ground-truth.py, ADR-079).
+COCO_FROM_MP = [0, 2, 5, 7, 8, 11, 12, 13, 14, 15, 16, 23, 24, 25, 26, 27, 28]
+COCO_NAMES = ["nose","l_eye","r_eye","l_ear","r_ear","l_sho","r_sho","l_elb",
+ "r_elb","l_wri","r_wri","l_hip","r_hip","l_knee","r_knee","l_ank","r_ank"]
+
+# ---- shared state between the CSI (async) thread and the camera (sync) loop ----
+_latest_csi = {"t": 0.0, "frame": None}
+_csi_lock = threading.Lock()
+_stop = threading.Event()
+
+
+def csi_thread(ws_url: str):
+ """Background thread: keep the most recent LIVE csi frame in _latest_csi."""
+ async def run():
+ while not _stop.is_set():
+ try:
+ async with websockets.connect(ws_url, open_timeout=8, ping_interval=20) as ws:
+ while not _stop.is_set():
+ msg = await asyncio.wait_for(ws.recv(), timeout=8)
+ d = json.loads(msg)
+ with _csi_lock:
+ _latest_csi["t"] = time.time()
+ _latest_csi["frame"] = d
+ except Exception as e:
+ print(f"[csi] reconnect ({e})", flush=True)
+ await asyncio.sleep(1.0)
+ asyncio.new_event_loop().run_until_complete(run())
+
+
+def csi_vector(frame: dict):
+ """Flatten a csi frame to a fixed-length input vector: features + field."""
+ f = frame.get("features", {}) or {}
+ feats = [f.get("mean_rssi", 0.0), f.get("variance", 0.0),
+ f.get("motion_band_power", 0.0), f.get("breathing_band_power", 0.0)]
+ # per-node mean_rssi/variance/motion for up to the 2 nodes (9, 13)
+ pernode = {nf.get("node_id"): (nf.get("features") or {}) for nf in (frame.get("node_features") or [])}
+ for nid in (9, 13):
+ nf = pernode.get(nid, {})
+ feats += [nf.get("mean_rssi", 0.0), nf.get("variance", 0.0), nf.get("motion_band_power", 0.0)]
+ field = (frame.get("signal_field", {}) or {}).get("values") or []
+ field = (field + [0.0] * 400)[:400]
+ return feats + field # 4 + 6 + 400 = 410-d
+
+
+def main():
+ ap = argparse.ArgumentParser(description="WiFlow camera-supervised CSI<->pose capture (ADR-180).")
+ ap.add_argument("--ws", default="ws://localhost:8765/ws/sensing")
+ ap.add_argument("--cam", type=int, default=0)
+ ap.add_argument("--out", default=os.path.expanduser("~/wiflow-room/dataset.jsonl"))
+ ap.add_argument("--seconds", type=int, default=180)
+ ap.add_argument("--min-vis", type=float, default=0.5, help="min mean landmark visibility to accept a pose label")
+ ap.add_argument("--max-skew-ms", type=float, default=150, help="max csi/pose time skew to pair")
+ ap.add_argument("--require-esp32", action="store_true", default=True,
+ help="only pair when csi source==esp32 (real). Default on.")
+ args = ap.parse_args()
+
+ os.makedirs(os.path.dirname(args.out), exist_ok=True)
+ th = threading.Thread(target=csi_thread, args=(args.ws,), daemon=True)
+ th.start()
+
+ cap = cv2.VideoCapture(args.cam)
+ if not cap.isOpened():
+ print(f"ERROR: cannot open camera {args.cam}", file=sys.stderr); sys.exit(2)
+ W = int(cap.get(cv2.CAP_PROP_FRAME_WIDTH)) or 640
+ H = int(cap.get(cv2.CAP_PROP_FRAME_HEIGHT)) or 480
+ model_path = ensure_model(os.path.expanduser("~/wiflow-room/pose_landmarker_lite.task"))
+ landmarker = PoseLandmarker.create_from_options(PoseLandmarkerOptions(
+ base_options=BaseOptions(model_asset_path=model_path),
+ running_mode=RunningMode.IMAGE, min_pose_detection_confidence=0.5))
+
+ n_pairs = 0; n_nopose = 0; n_nocsi = 0; n_skew = 0; n_sim = 0
+ t0 = time.time()
+ print(f"[capture] camera {args.cam} {W}x{H} -> {args.out} for {args.seconds}s")
+ print("[capture] stand in view AND in the CSI field; move/walk so poses vary. Ctrl-C to stop.")
+ with open(args.out, "a") as out:
+ try:
+ while time.time() - t0 < args.seconds:
+ ok, frame = cap.read()
+ if not ok:
+ continue
+ now = time.time()
+ rgb = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
+ res = landmarker.detect(mp.Image(image_format=mp.ImageFormat.SRGB, data=rgb))
+ if not res.pose_landmarks:
+ n_nopose += 1; continue
+ lm = res.pose_landmarks[0]
+ kps = [[lm[i].x, lm[i].y, lm[i].visibility] for i in COCO_FROM_MP]
+ vis = float(np.mean([k[2] for k in kps]))
+ if vis < args.min_vis:
+ n_nopose += 1; continue
+ with _csi_lock:
+ ct = _latest_csi["t"]; cf = _latest_csi["frame"]
+ if cf is None:
+ n_nocsi += 1; continue
+ if (now - ct) * 1000.0 > args.max_skew_ms:
+ n_skew += 1; continue
+ if args.require_esp32 and cf.get("source") != "esp32":
+ n_sim += 1; continue
+ rec = {"t": now, "vis": round(vis, 3),
+ "kps": [[round(x, 4), round(y, 4), round(v, 3)] for x, y, v in kps],
+ "csi": csi_vector(cf),
+ "src": cf.get("source"),
+ "nodes": sorted(n.get("node_id") for n in cf.get("nodes", []) if n.get("node_id") is not None)}
+ out.write(json.dumps(rec) + "\n")
+ n_pairs += 1
+ if n_pairs % 30 == 0:
+ out.flush()
+ el = int(now - t0)
+ print(f"[capture] t+{el:3d}s pairs={n_pairs} (skip: nopose={n_nopose} nocsi={n_nocsi} skew={n_skew} sim={n_sim})", flush=True)
+ except KeyboardInterrupt:
+ print("\n[capture] stopped by user")
+ _stop.set(); cap.release()
+ print(f"[capture] DONE. wrote {n_pairs} paired samples to {args.out}")
+ print(f"[capture] skipped: no-pose={n_nopose} no-csi={n_nocsi} skew={n_skew} simulated={n_sim}")
+ if n_pairs == 0:
+ print("[capture] WARNING: 0 pairs — check camera sees you AND csi source==esp32 (live).")
+
+
+if __name__ == "__main__":
+ main()
diff --git a/examples/through-wall/wiflow_infer.py b/examples/through-wall/wiflow_infer.py
new file mode 100644
index 00000000..aedce4e0
--- /dev/null
+++ b/examples/through-wall/wiflow_infer.py
@@ -0,0 +1,92 @@
+#!/usr/bin/env python3
+"""Live CSI->pose inference bridge (ADR-180).
+
+Runs on the box with the live CSI. Loads the camera-supervised model (numpy,
+no torch needed), subscribes to /ws/sensing, runs a forward pass per frame, and
+broadcasts the predicted 17-keypoint pose to HTML clients on ws://:8770/pose.
+
+ python wiflow_infer.py --model model/model.npz \
+ --in ws://localhost:8765/ws/sensing --port 8770
+"""
+import argparse, asyncio, json, os
+import numpy as np
+import websockets
+
+# COCO skeleton edges (for the client; sent once in 'meta')
+EDGES = [[5,7],[7,9],[6,8],[8,10],[5,6],[11,12],[5,11],[6,12],
+ [11,13],[13,15],[12,14],[14,16],[0,1],[0,2],[1,3],[2,4],[0,5],[0,6]]
+
+def csi_vector(frame):
+ f = frame.get("features", {}) or {}
+ feats = [f.get("mean_rssi",0.0), f.get("variance",0.0),
+ f.get("motion_band_power",0.0), f.get("breathing_band_power",0.0)]
+ pernode = {nf.get("node_id"): (nf.get("features") or {}) for nf in (frame.get("node_features") or [])}
+ for nid in (9,13):
+ nf = pernode.get(nid,{}); feats += [nf.get("mean_rssi",0.0), nf.get("variance",0.0), nf.get("motion_band_power",0.0)]
+ field = (frame.get("signal_field",{}) or {}).get("values") or []
+ field = (field + [0.0]*400)[:400]
+ return np.array(feats + field, np.float32)
+
+class Model:
+ def __init__(self, path):
+ z = np.load(path)
+ self.mu, self.sd = z["mu"], z["sd"]
+ self.W = [z["net_0_weight"], z["net_3_weight"], z["net_6_weight"], z["net_8_weight"]]
+ self.b = [z["net_0_bias"], z["net_3_bias"], z["net_6_bias"], z["net_8_bias"]]
+ def __call__(self, x):
+ h = (x - self.mu) / self.sd
+ for i in range(3):
+ h = np.maximum(0.0, h @ self.W[i].T + self.b[i]) # Linear+ReLU
+ out = 1.0/(1.0+np.exp(-(h @ self.W[3].T + self.b[3]))) # Linear+Sigmoid -> 34
+ return out.reshape(17,2)
+
+CLIENTS = set()
+LATEST = {"pose": None}
+
+async def serve_client(ws):
+ CLIENTS.add(ws)
+ try:
+ await ws.send(json.dumps({"type":"meta","edges":EDGES}))
+ async for _ in ws: # client is read-only; just keep alive
+ pass
+ except Exception:
+ pass
+ finally:
+ CLIENTS.discard(ws)
+
+async def infer_loop(model, in_url):
+ while True:
+ try:
+ async with websockets.connect(in_url, open_timeout=8, ping_interval=20) as ws:
+ async for msg in ws:
+ d = json.loads(msg)
+ kp = model(csi_vector(d))
+ cls = d.get("classification",{})
+ payload = {"type":"pose","src":d.get("source"),
+ "presence":bool(cls.get("presence")),
+ "motion":(d.get("features",{}) or {}).get("motion_band_power"),
+ "kps":[[round(float(x),4),round(float(y),4)] for x,y in kp],
+ "nodes":sorted(n.get("node_id") for n in d.get("nodes",[]) if n.get("node_id") is not None)}
+ LATEST["pose"]=payload
+ if CLIENTS:
+ dead=[]
+ for c in list(CLIENTS):
+ try: await c.send(json.dumps(payload))
+ except Exception: dead.append(c)
+ for c in dead: CLIENTS.discard(c)
+ except Exception as e:
+ print(f"[infer] reconnect ({e})", flush=True); await asyncio.sleep(1.0)
+
+async def main():
+ ap = argparse.ArgumentParser()
+ ap.add_argument("--model", default=os.path.join(os.path.dirname(__file__),"model","model.npz"))
+ ap.add_argument("--in", dest="in_url", default="ws://localhost:8765/ws/sensing")
+ ap.add_argument("--port", type=int, default=8770)
+ args = ap.parse_args()
+ model = Model(args.model)
+ print(f"[infer] model {args.model} loaded; serving predicted poses on ws://0.0.0.0:{args.port}/pose")
+ async with websockets.serve(serve_client, "0.0.0.0", args.port):
+ await infer_loop(model, args.in_url)
+
+if __name__ == "__main__":
+ asyncio.run(main())
diff --git a/examples/through-wall/wiflow_train.py b/examples/through-wall/wiflow_train.py
new file mode 100644
index 00000000..7d0bd809
--- /dev/null
+++ b/examples/through-wall/wiflow_train.py
@@ -0,0 +1,102 @@
+#!/usr/bin/env python3
+"""Train a CSI->pose model on the camera-supervised dataset (ADR-079/180).
+
+Input : 410-d CSI vector (4 global feats + 6 per-node + 400 signal-field).
+Target : 17 COCO keypoints (x,y), normalized 0..1 from the camera (ground truth).
+Reports HONEST held-out PCK@k + MPJPE on a chronological val split (the last
+20% of the session — never trained on), so the number is not leaked.
+
+Usage (ruvultra venv):
+ python wiflow_train.py --data ~/wiflow-room/dataset.jsonl --out ~/wiflow-room/model.pt
+"""
+import argparse, json, math, os, sys
+import numpy as np
+import torch, torch.nn as nn
+
+
+def load(path):
+ X, Y, V = [], [], []
+ with open(path) as f:
+ for line in f:
+ r = json.loads(line)
+ X.append(r["csi"]) # 410
+ kp = r["kps"] # 17 x [x,y,vis]
+ Y.append([c for k in kp for c in (k[0], k[1])]) # 34
+ V.append([k[2] for k in kp]) # 17 visibilities
+ return np.array(X, np.float32), np.array(Y, np.float32), np.array(V, np.float32)
+
+
+class Net(nn.Module):
+ def __init__(self, din, dout):
+ super().__init__()
+ self.net = nn.Sequential(
+ nn.Linear(din, 512), nn.ReLU(), nn.Dropout(0.3),
+ nn.Linear(512, 256), nn.ReLU(), nn.Dropout(0.3),
+ nn.Linear(256, 128), nn.ReLU(),
+ nn.Linear(128, dout), nn.Sigmoid()) # coords in 0..1
+ def forward(self, x): return self.net(x)
+
+
+def pck(pred, gt, vis, thr):
+ # pred/gt: [N,34] -> [N,17,2]; PCK@thr in normalized image units, visible kps only
+ p = pred.reshape(-1, 17, 2); g = gt.reshape(-1, 17, 2)
+ d = np.linalg.norm(p - g, axis=2) # [N,17]
+ m = vis > 0.5
+ return float((d[m] < thr).mean()) if m.any() else 0.0, float(d[m].mean()) if m.any() else float("nan")
+
+
+def main():
+ ap = argparse.ArgumentParser()
+ ap.add_argument("--data", required=True)
+ ap.add_argument("--out", default=os.path.expanduser("~/wiflow-room/model.pt"))
+ ap.add_argument("--epochs", type=int, default=300)
+ ap.add_argument("--bs", type=int, default=64)
+ args = ap.parse_args()
+
+ X, Y, V = load(args.data)
+ n = len(X)
+ print(f"[train] {n} samples, X={X.shape} Y={Y.shape}")
+ if n < 200:
+ print("[train] too few samples"); sys.exit(2)
+
+ # chronological split (NOT shuffled) so val is a held-out time segment -> honest
+ cut = int(n * 0.8)
+ mu, sd = X[:cut].mean(0), X[:cut].std(0) + 1e-6 # standardize on train only
+ Xn = (X - mu) / sd
+ dev = "cuda" if torch.cuda.is_available() else "cpu"
+ Xtr = torch.tensor(Xn[:cut]).to(dev); Ytr = torch.tensor(Y[:cut]).to(dev)
+ Xva = torch.tensor(Xn[cut:]).to(dev); Yva = Y[cut:]; Vva = V[cut:]
+
+ # mean-pose baseline (predict the train-mean pose for everything) — the bar to beat
+ mean_pose = Y[:cut].mean(0)
+ base_pck, base_mpjpe = pck(np.tile(mean_pose, (len(Yva), 1)), Yva, Vva, 0.10)
+
+ net = Net(X.shape[1], Y.shape[1]).to(dev)
+ opt = torch.optim.Adam(net.parameters(), lr=1e-3, weight_decay=1e-4)
+ lossf = nn.MSELoss()
+ best = (1e9, None)
+ for ep in range(args.epochs):
+ net.train(); perm = torch.randperm(len(Xtr), device=dev)
+ for i in range(0, len(Xtr), args.bs):
+ idx = perm[i:i+args.bs]
+ opt.zero_grad(); out = net(Xtr[idx]); loss = lossf(out, Ytr[idx]); loss.backward(); opt.step()
+ if (ep + 1) % 20 == 0 or ep == args.epochs - 1:
+ net.eval()
+ with torch.no_grad(): pv = net(Xva).cpu().numpy()
+ p10, mpj = pck(pv, Yva, Vva, 0.10); p05, _ = pck(pv, Yva, Vva, 0.05)
+ vloss = float(((pv - Yva) ** 2).mean())
+ print(f"[train] ep{ep+1:3d} val_mse={vloss:.4f} PCK@0.10={p10*100:.1f}% PCK@0.05={p05*100:.1f}% MPJPE={mpj:.4f}")
+ if vloss < best[0]: best = (vloss, {"sd": net.state_dict(), "p10": p10, "p05": p05, "mpj": mpj})
+
+ torch.save({"model": best[1]["sd"], "mu": mu, "sd": sd, "din": X.shape[1]}, args.out)
+ print("\n==================== HONEST RESULT (held-out 20%, never trained) ====================")
+ print(f" MEAN-POSE BASELINE : PCK@0.10 = {base_pck*100:.1f}% MPJPE = {base_mpjpe:.4f} (the bar to beat)")
+ print(f" CSI->POSE MODEL : PCK@0.10 = {best[1]['p10']*100:.1f}% PCK@0.05 = {best[1]['p05']*100:.1f}% MPJPE = {best[1]['mpj']:.4f}")
+ delta = (best[1]['p10'] - base_pck) * 100
+ print(f" VERDICT: model {'BEATS' if delta>1 else 'does NOT beat'} mean-pose baseline by {delta:+.1f} pp "
+ f"-> {'real CSI->pose signal' if delta>1 else 'NO usable CSI->pose signal (honest negative)'}")
+ print(f" saved -> {args.out}")
+
+
+if __name__ == "__main__":
+ main()