Update Dockerfile

Dockerfile

@@ -1,28 +1,64 @@
-# Use an official Python runtime
-FROM python:3.10
+# Dockerfile
+
+
+FROM python:3.10-slim
 
-# Set the working directory
 WORKDIR /app
 
-# --- UPDATED: Use libgl1 instead of the obsolete libgl1-mesa-glx ---
-RUN apt-get update && apt-get install -y \
-    libgl1 \
-    libglib2.0-0 \
+# ── System deps (OpenCV headless needs libGL) ────────────────────
+RUN apt-get update && apt-get install -y --no-install-recommends \
+        libgl1 libglib2.0-0 libgomp1 git \
     && rm -rf /var/lib/apt/lists/*
 
-# Copy requirements and install them
+# ── Python deps ───────────────────────────────────────────────────
 COPY requirements.txt .
-RUN pip install --no-cache-dir -r requirements.txt
+RUN pip install --no-cache-dir --compile -r requirements.txt
 
-# Copy the rest of the backend code
+# ── Copy application code ────────────────────────────────────────
 COPY . .
 
-# Create the temp directory and give it permission
-RUN mkdir -p temp_uploads
-RUN chmod -R 777 temp_uploads
+RUN mkdir -p temp_uploads saved_images && chmod -R 777 temp_uploads saved_images
+
+# ── Pre-download all AI models at BUILD time ─────────────────────
+# This bakes the weights into the Docker image layer.
+# Cold-start on HF Spaces goes from ~3-5 min → ~10 sec.
+# Remove this block if your image size budget is tight (<5 GB limit on free HF).
+RUN python - <<'EOF'
+from transformers import AutoProcessor, AutoModel, AutoImageProcessor
+from ultralytics import YOLO
+from deepface import DeepFace
+import numpy as np
+
+print("Pre-downloading SigLIP …")
+AutoProcessor.from_pretrained("google/siglip-base-patch16-224", use_fast=True)
+AutoModel.from_pretrained("google/siglip-base-patch16-224")
+
+print("Pre-downloading DINOv2 …")
+AutoImageProcessor.from_pretrained("facebook/dinov2-base")
+AutoModel.from_pretrained("facebook/dinov2-base")
+
+print("Pre-downloading YOLO …")
+YOLO("yolo11n.pt")
+
+print("Pre-downloading GhostFaceNet + RetinaFace …")
+dummy = np.zeros((112, 112, 3), dtype=np.uint8)
+try:
+    DeepFace.represent(img_path=dummy, model_name="GhostFaceNet",
+                       detector_backend="retinaface", enforce_detection=False)
+except Exception:
+    pass   # first run just downloads weights; inference error is fine here
+
+print("✅ All models cached in image layer")
+EOF
 
-# Hugging Face port
 EXPOSE 7860
 
-# Start the FastAPI server
-CMD ["uvicorn", "main:app", "--host", "0.0.0.0", "--port", "7860"]
+# ── Two uvicorn workers for true parallelism ─────────────────────
+# WEB_CONCURRENCY can be overridden via HF Space env vars
+ENV WEB_CONCURRENCY=2
+
+CMD uvicorn main:app \
+        --host 0.0.0.0 \
+        --port 7860 \
+        --workers ${WEB_CONCURRENCY} \
+        --timeout-keep-alive 75