feat(ai): dual-model face detection with NMS deduplication

Run both short-range and full-range MediaPipe models and merge results,
then apply non-maximum suppression to remove duplicate bounding boxes.
Fixes missed faces in group photos where the single-model loop exited
early after the first positive detection.

packages/ai/python/detect_faces.py

@@ -30,12 +30,48 @@ def _ensure_face_detect_model():
     return _LOCAL_MODEL_PATH
 
 
+def _iou(a, b):
+    """Compute intersection-over-union between two face boxes."""
+    ax2, ay2 = a["x"] + a["w"], a["y"] + a["h"]
+    bx2, by2 = b["x"] + b["w"], b["y"] + b["h"]
+    inter_w = max(0, min(ax2, bx2) - max(a["x"], b["x"]))
+    inter_h = max(0, min(ay2, by2) - max(a["y"], b["y"]))
+    inter = inter_w * inter_h
+    union = a["w"] * a["h"] + b["w"] * b["h"] - inter
+    return inter / union if union > 0 else 0.0
+
+
+def _nms_faces(faces, iou_threshold=0.4):
+    """Remove duplicate detections using greedy non-maximum suppression."""
+    if len(faces) <= 1:
+        return faces
+    kept = []
+    used = [False] * len(faces)
+    for i in range(len(faces)):
+        if used[i]:
+            continue
+        kept.append(faces[i])
+        used[i] = True
+        for j in range(i + 1, len(faces)):
+            if not used[j] and _iou(faces[i], faces[j]) >= iou_threshold:
+                used[j] = True
+    return kept
+
+
 def _detect_with_solutions(img_array, min_confidence):
-    """Detect faces using legacy mp.solutions API (mediapipe < 0.10.30)."""
+    """Detect faces using legacy mp.solutions API (mediapipe < 0.10.30).
+
+    Runs both short-range (model 0) and full-range (model 1) detectors and
+    merges the results. Previously the loop broke on the first model that
+    found any face, so group photos where model 0 caught only 1-2 large
+    faces would never have the remaining faces scanned by model 1.
+    """
     import mediapipe as mp
 
     mp_face = mp.solutions.face_detection
-    results = None
+    ih, iw = img_array.shape[:2]
+    all_faces = []
+
     for model_sel in [0, 1]:
         detector = mp_face.FaceDetection(
             model_selection=model_sel,
@@ -43,24 +79,16 @@ def _detect_with_solutions(img_array, min_confidence):
         )
         results = detector.process(img_array)
         detector.close()
-        if results.detections:
-            break
+        for detection in (results.detections or []):
+            bbox = detection.location_data.relative_bounding_box
+            all_faces.append({
+                "x": int(bbox.xmin * iw),
+                "y": int(bbox.ymin * ih),
+                "w": int(bbox.width * iw),
+                "h": int(bbox.height * ih),
+            })
 
-    detections = results.detections or []
-    if not detections:
-        return []
-
-    ih, iw = img_array.shape[:2]
-    faces = []
-    for detection in detections:
-        bbox = detection.location_data.relative_bounding_box
-        faces.append({
-            "x": int(bbox.xmin * iw),
-            "y": int(bbox.ymin * ih),
-            "w": int(bbox.width * iw),
-            "h": int(bbox.height * ih),
-        })
-    return faces
+    return _nms_faces(all_faces)
 
 
 def _detect_with_tasks(img_array, min_confidence):