fix: Build failing

Dockerfile

@@ -1,6 +1,6 @@
 # Dockerfile — Enterprise Lens V3
-# InsightFace needs: g++, cmake, Cython, scikit-build, onnxruntime
-# All installed in correct order before insightface
+# InsightFace models download on first run (not at build time)
+# This avoids build timeout and network issues during Docker build
 
 FROM python:3.10-slim
 
@@ -14,7 +14,7 @@ RUN apt-get update && apt-get install -y --no-install-recommends \
         wget ca-certificates \
     && rm -rf /var/lib/apt/lists/*
 
-# ── Step 1: Core build tools (MUST be before insightface) ────────
+# ── Step 1: Build tools (MUST be before insightface) ─────────────
 RUN pip install --no-cache-dir \
         "numpy<2.0" \
         "setuptools>=65" \
@@ -26,7 +26,7 @@ RUN pip install --no-cache-dir \
 # ── Step 2: onnxruntime (MUST be before insightface) ─────────────
 RUN pip install --no-cache-dir onnxruntime
 
-# ── Step 3: insightface (now has all deps available) ─────────────
+# ── Step 3: insightface ───────────────────────────────────────────
 RUN pip install --no-cache-dir --prefer-binary insightface
 
 # ── Step 4: Remaining requirements ───────────────────────────────
@@ -37,7 +37,8 @@ RUN pip install --no-cache-dir --prefer-binary -r requirements.txt
 COPY . .
 RUN mkdir -p temp_uploads saved_images && chmod -R 777 temp_uploads saved_images
 
-# ── Pre-download all models at build time ─────────────────────────
+# ── Pre-download ONLY transformers + YOLO at build time ──────────
+# InsightFace models download on first startup (cached after that)
 RUN python - <<'EOF'
 import os
 os.environ["TRANSFORMERS_VERBOSITY"] = "error"
@@ -59,14 +60,7 @@ from ultralytics import YOLO
 YOLO("yolo11n-seg.pt")
 print("YOLO done")
 
-print("Pre-downloading InsightFace buffalo_sc...")
-from insightface.app import FaceAnalysis
-app = FaceAnalysis(name="buffalo_sc", providers=["CPUExecutionProvider"])
-app.prepare(ctx_id=-1, det_size=(640, 640))
-del app
-print("InsightFace done")
-
-print("All V3 models cached!")
+print("Build complete! InsightFace models download on first startup.")
 EOF
 
 EXPOSE 7860

src/models.py

@@ -19,6 +19,7 @@ import io
 
 import cv2
 import numpy as np
+import threading
 import torch
 import torch.nn.functional as F
 from PIL import Image
@@ -113,14 +114,12 @@ class AIModelManager:
                 # buffalo_sc = small+fast model (CPU optimised)
                 # buffalo_l  = large+accurate (use if GPU available)
                 model_name = "buffalo_l" if self.device == "cuda" else "buffalo_sc"
-                self.face_app = FaceAnalysis(
-                    name=model_name,
-                    providers=["CUDAExecutionProvider"] if self.device == "cuda"
-                              else ["CPUExecutionProvider"],
-                )
+                self.face_app = FaceAnalysis(name=model_name)
                 # det_size controls detection resolution — larger = finds smaller faces
-                self.face_app.prepare(ctx_id=0 if self.device == "cuda" else -1,
-                                      det_size=(640, 640))
+                self.face_app.prepare(
+                    ctx_id=0 if self.device == "cuda" else -1,
+                    det_size=(640, 640),
+                )
                 print(f"✅ InsightFace ({model_name}) loaded — ArcFace face lane active")
             except Exception as e:
                 print(f"⚠️  InsightFace init failed: {e} — face lane disabled")
@@ -130,6 +129,9 @@ class AIModelManager:
 
         self._cache          = {}
         self._cache_maxsize  = 128
+        # InsightFace ONNX runtime is NOT thread-safe
+        # This lock ensures only one inference runs at a time
+        self._face_lock      = threading.Lock()
         print("✅ Models ready!")
 
     # ── Object Lane batched embedding ────────────────────────────
@@ -183,7 +185,8 @@ class AIModelManager:
             else:
                 bgr = img_np.copy()
 
-            faces = self.face_app.get(bgr)
+            with self._face_lock:
+                faces = self.face_app.get(bgr)
             results = []
 
             for idx, face in enumerate(faces):