Fix AdaptFormer on production: pin protobuf and expose model health in /health

Dockerfile

@@ -19,14 +19,15 @@ WORKDIR /app
 
 # Build-time info + cache-bust:
 # Changing APP_BUILD forces Docker to re-run subsequent layers (including pip install).
-ARG APP_BUILD=23
+ARG APP_BUILD=24
 ENV APP_BUILD=${APP_BUILD}
 RUN echo "Docker build start: APP_BUILD=${APP_BUILD}" && python -V
 
 # Install Python dependencies
 COPY requirements.txt .
 RUN pip install --no-cache-dir --disable-pip-version-check --default-timeout=300 -U pip setuptools wheel
-RUN pip install --no-cache-dir --disable-pip-version-check --default-timeout=300 --prefer-binary -r requirements.txt -v
+RUN pip install --no-cache-dir --disable-pip-version-check --default-timeout=300 --prefer-binary -r requirements.txt -v \
+    && python -c "import google.protobuf; from transformers import AutoImageProcessor; print('protobuf', google.protobuf.__version__, 'transformers ok')"
 
 # Pre-download the AdaptFormer model so cold starts are instant
 ENV HF_HOME=/app/.hf_cache

app/main.py

@@ -70,14 +70,22 @@ except Exception as e:
     import logging
     logging.getLogger("uvicorn.error").warning("Startup migration skipped: %s", e)
 
-app = FastAPI(title="AI Change Detection", version="2.2.0")
+app = FastAPI(title="AI Change Detection", version="2.2.1")
 
 
 @app.get("/health")
 def health():
-    """Lightweight health check so Hugging Face can mark the Space as running quickly."""
+    """Health check + AdaptFormer model status (HF Spaces + diagnostics)."""
     from datetime import datetime
-    return {"status": "ok", "version": "2.2.0", "server_time_ist": _isoformat_ist(datetime.now(timezone.utc))}
+    from .model_inference import get_model_status
+
+    model = get_model_status()
+    return {
+        "status": "ok" if model.get("available") else "degraded",
+        "version": "2.2.1",
+        "server_time_ist": _isoformat_ist(datetime.now(timezone.utc)),
+        "adaptFormer": model,
+    }
 
 
 @app.on_event("startup")

app/model_inference.py

@@ -22,6 +22,7 @@ _MODEL_ID = "deepang/adaptformer-LEVIR-CD"
 _TILE_SIZE = 256  # LEVIR-CD native patch size
 _AVAILABLE = None
 _LOAD_FAILED = False
+_LOAD_ERROR: str | None = None
 
 
 def _try_import():
@@ -34,7 +35,7 @@ def _try_import():
 
 
 def _load_model():
-    global _MODEL, _PROCESSOR, _DEVICE, _AVAILABLE, _LOAD_FAILED
+    global _MODEL, _PROCESSOR, _DEVICE, _AVAILABLE, _LOAD_FAILED, _LOAD_ERROR
     if _MODEL is not None:
         return _MODEL, _PROCESSOR
     if _LOAD_FAILED:
@@ -56,10 +57,12 @@ def _load_model():
         _MODEL.to(_DEVICE)
         _MODEL.eval()
         _AVAILABLE = True
+        _LOAD_ERROR = None
         logger.info("AdaptFormer loaded on %s", _DEVICE)
     except Exception as exc:
         _LOAD_FAILED = True
         _AVAILABLE = False
+        _LOAD_ERROR = str(exc)
         logger.error("AdaptFormer load failed: %s", exc)
         raise
     return _MODEL, _PROCESSOR
@@ -81,15 +84,38 @@ def is_model_available():
 
 def preload_model():
     """Warm-load AdaptFormer at app startup (best-effort)."""
+    global _LOAD_ERROR
     try:
         _load_model()
         logger.info("AdaptFormer preload complete")
         return True
     except Exception as exc:
+        _LOAD_ERROR = str(exc)
         logger.warning("AdaptFormer preload skipped: %s", exc)
         return False
 
 
+def get_model_status() -> dict:
+    """Status for /health — shows whether AI detection or classical fallback is active."""
+    if _AVAILABLE is True:
+        mode = "adaptformer_gated_fusion"
+        available = True
+    elif _LOAD_FAILED:
+        mode = "classical_fallback"
+        available = False
+    else:
+        available = is_model_available()
+        mode = "adaptformer_gated_fusion" if available else "classical_fallback"
+
+    return {
+        "modelId": _MODEL_ID,
+        "available": available,
+        "detectionMode": mode,
+        "device": str(_DEVICE) if _DEVICE is not None else None,
+        "error": _LOAD_ERROR,
+    }
+
+
 def predict_change_mask(img1, img2, threshold=0.5):
     """
     Run AdaptFormer inference on two RGB numpy arrays (H, W, 3).

requirements.txt

@@ -11,7 +11,8 @@ python-jose[cryptography]>=3.3.0
 passlib[bcrypt]>=1.7.4
 bcrypt==4.0.1
 pillow>=10.0.0
-numpy>=1.24.0
+numpy>=1.26,<2
 opencv-python-headless>=4.8.0
 scikit-learn>=1.3.0
 requests>=2.28.0
+protobuf>=5.28.0,<6