Upload wiflow-v1/MODEL_CARD.md with huggingface_hub

wiflow-v1/MODEL_CARD.md

@@ -0,0 +1,31 @@
+# WiFlow v1 — WiFi CSI Pose Estimation Model
+
+Camera-supervised WiFlow pose model trained on real ESP32 CSI + MediaPipe ground truth.
+
+## Metrics
+- **PCK@20: 92.9%** (17 COCO keypoints)
+- Eval loss: 0.082
+- Bone constraint loss: 0.008
+- Parameters: 186,946 (974 KB)
+
+## Architecture
+- TCN: 2 dilated causal conv blocks (k=3, d=[1,2])
+- Input: 35 subcarriers x 20 time steps (ruvector-solver reduced from 70)
+- Output: 17 COCO keypoints [x, y] in [0, 1]
+- Scale: lite (189K params)
+
+## Training
+- 345 paired samples (5 min capture)
+- ESP32-S3 CSI: 7,000 frames at 23fps
+- Mac camera: 6,470 frames at 22fps via MediaPipe PoseLandmarker
+- 50 epochs (33 supervised + 17 refinement)
+- ruvector optimizations: O6 subcarrier selection, O7 attention, O8 Stoer-Wagner min-cut, O9 multi-SPSA
+
+## Usage
+```javascript
+const model = JSON.parse(fs.readFileSync('wiflow-v1.json'));
+// Load into WiFlowSupervisedModel and call forward(csiInput)
+```
+
+## License
+Apache-2.0