Add error handling fix

model-lab/app/main.py

@@ -6,15 +6,21 @@
 from pathlib import Path
 from fastapi.middleware.cors import CORSMiddleware
 from dotenv import load_dotenv
+
 load_dotenv()
 
 app = FastAPI(title="Face Sentiment API")
 
-# path to the model
+# Path to the model (fixed typo: onxx_models → onnx_models)
 BASE_DIR = Path(__file__).resolve().parent.parent
-MODEL_PATH = BASE_DIR / "models"/ "onxx_models" / "emotion-ferplus-8.onnx"
+MODEL_PATH = BASE_DIR / "models" / "onnx_models" / "emotion-ferplus-8.onnx"
+if not MODEL_PATH.exists():
+    raise FileNotFoundError(f"Model file not found: {MODEL_PATH}")
 
+# Validate CORS origins properly
 ALLOWED_ORIGINS = os.getenv("CORS_ALLOWED_ORIGINS", "http://localhost:3000").split(",")
+if not ALLOWED_ORIGINS or ALLOWED_ORIGINS == [""]:
+    raise RuntimeError("No allowed origins configured in CORS_ALLOWED_ORIGINS")
 
 app.add_middleware(
     CORSMiddleware,
@@ -23,16 +29,31 @@
     allow_headers=["Content-Type"],
 )
 
+# ✅ Pre-load both model options at startup to avoid race conditions
+emotion_models = {
+    1: Model(model_path=MODEL_PATH, model_option=1),  # HuggingFace ViT
+    2: Model(model_path=MODEL_PATH, model_option=2)   # ONNX with CV2
+}
+
 @app.post("/predict")
 async def predict(file: UploadFile = File(...), model_option: int = Form(1)):
     if not file.content_type.startswith("image/"):
         raise HTTPException(status_code=400, detail="File must be an image")
+
     file_bytes = await file.read()
     try:
         image = Image.open(io.BytesIO(file_bytes))
     except (OSError, ValueError) as err:
         raise HTTPException(status_code=400, detail="Invalid or corrupted image file") from err
-    emotion_model  = Model(model_path=MODEL_PATH, model_option=model_option)
-    emotion, prob = emotion_model.predict(pil_image=image)
-
-    return {"emotion": emotion, "probabilities": prob}
+
+    # ✅ Validate model_option before using
+    if model_option not in emotion_models:
+        raise HTTPException(status_code=400, detail=f"Invalid model_option: {model_option}")
+
+    try:
+        # ✅ Select the correct preloaded model without mutating shared state
+        emotion, prob = emotion_models[model_option].predict(pil_image=image)
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=f"Model inference failed: {e}") from e
+
+    return {"emotion": emotion, "probabilities": prob}