Create app.py

app.py

@@ -0,0 +1,50 @@
+import gradio as gr
+import cv2, numpy as np
+import mediapipe as mp
+
+# --- your existing imports / functions ---
+# from your_module import au_measures, stress_score_from_features   # if you split them out
+
+mp_face_mesh = mp.solutions.face_mesh
+face_mesh = mp_face_mesh.FaceMesh(
+    min_detection_confidence=0.5,
+    min_tracking_confidence=0.5,
+    refine_landmarks=True
+)
+
+def compute_stress_score_bgr(frame_bgr: np.ndarray) -> dict:
+    """Minimal example: run FaceMesh + your AU→stress logic and return a dict."""
+    h, w = frame_bgr.shape[:2]
+    rgb = cv2.cvtColor(frame_bgr, cv2.COLOR_BGR2RGB)
+    res = face_mesh.process(rgb)
+
+    if not res.multi_face_landmarks:
+        return {"status": "no_face", "stress_score": None, "label": "NO FACE"}
+
+    lms = res.multi_face_landmarks[0].landmark
+
+    # ---- plug in your real AU code here ----
+    # F = au_measures(lms, w, h)
+    # score = stress_score_from_features(F)
+    score = 42.0  # <--- placeholder; replace with your real computation
+    label = "STRESSED" if score >= 55 else ("POSSIBLY STRESSED" if score >= 25 else "NOT STRESSED")
+
+    return {"status": "ok", "stress_score": round(score, 2), "label": label}
+
+def predict(frame: np.ndarray):
+    """
+    Gradio streaming callback receives a single frame (BGR numpy array).
+    Return any JSON-serializable object. You can also overlay results and return a video frame.
+    """
+    out = compute_stress_score_bgr(frame)
+    return out
+
+demo = gr.Interface(
+    fn=predict,
+    inputs=gr.Video(source="webcam", streaming=True),   # webcam in the browser → streamed to Python
+    outputs=gr.JSON(label="Result"),
+    title="StressDetection V1 (MediaPipe)"
+)
+
+if __name__ == "__main__":
+    demo.launch()