feat: add Qdrant _faces collection embedding push

- Add qdrant_faces.py utility module for _faces collection operations
- Modify face_processor.py to push embeddings to Qdrant (CoreML extraction re-enabled)
- Modify store_traced_faces.py to update trace_id in Qdrant after face tracking
- Collection schema: 512D vectors, Cosine distance, fixed name '_faces'
- Payload: file_uuid, frame, trace_id, bbox, confidence, identity_id/uuid, stranger_id
- Batch size: 100 (default), configurable via QDRANT_BATCH_SIZE env var
- Error handling: face_processor.py exits with error if Qdrant push fails

scripts/face_processor.py

@@ -30,7 +30,9 @@ from pathlib import Path
 import coremltools as ct
 
 sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+sys.path.insert(0, os.path.join(os.path.dirname(os.path.abspath(__file__)), "utils"))
 from redis_publisher import RedisPublisher
+from qdrant_faces import push_face_embeddings_batch
 
 SCRIPT_DIR = os.path.dirname(os.path.abspath(__file__))
 SWIFT_BIN = os.path.join(SCRIPT_DIR, "swift_processors", ".build", "debug", "swift_face_pose")
@@ -199,6 +201,7 @@ class FaceProcessorVision:
         embed_count = 0
         total_face_count = 0
         last_pct = -1
+        all_embeddings = []  # Collect embeddings for Qdrant push
 
         for frame_info in frames:
             frame_num = frame_info["frame"]
@@ -225,11 +228,18 @@ class FaceProcessorVision:
                 if face_img.size == 0:
                     continue
 
-                # CoreML embedding - TODO: push to Qdrant _faces collection instead
-                # emb = self.extract_face_embedding(face_img)
-                emb = None
+                # CoreML embedding - push to Qdrant _faces collection
+                emb = self.extract_face_embedding(face_img)
                 if emb is not None:
                     embed_count += 1
+                    # Collect for batch Qdrant push
+                    all_embeddings.append({
+                        "frame": frame_num,
+                        "trace_id": 0,  # Initial, updated by face_tracker
+                        "bbox": {"x": x, "y": y, "width": w, "height": h},
+                        "confidence": face.get("confidence", 0.5),
+                        "embedding": emb,
+                    })
 
                 # Pose classification
                 pose_info = face.get("pose", {})
@@ -292,6 +302,18 @@ class FaceProcessorVision:
         with open(self.output_path, "w") as f:
             json.dump(output, f, indent=2, ensure_ascii=False)
 
+        # Push embeddings to Qdrant _faces collection
+        if all_embeddings:
+            try:
+                pushed = push_face_embeddings_batch(self.uuid, all_embeddings, self.publisher)
+                if pushed != len(all_embeddings):
+                    raise RuntimeError(
+                        f"Qdrant push incomplete: {pushed}/{len(all_embeddings)} embeddings pushed"
+                    )
+            except Exception as e:
+                print(f"[FACE_V2] ERROR: Qdrant push failed: {e}", file=sys.stderr)
+                raise RuntimeError(f"Qdrant push failed: {e}")
+
         elapsed = time.time() - t0
         print(f"[FACE_V2] Done: {len(frames_list)} frames, {embed_count} embeddings, {elapsed:.0f}s")