Update app.py with latest changes

app.py

@@ -16,12 +16,17 @@ from PIL import Image
 import torch
 import torch.nn.functional as F
 from transformers import CLIPProcessor, CLIPModel
+from transformers import pipeline as hf_pipeline
 
 # --- Configuration ---
 # LITE varijanta: CLIP zero-shot klasifikacija nad Food-101 labelama (CPU-friendly)
 # Zadano koristi ViT-L/14 model; može se promijeniti preko env varijable MODEL_NAME
 MODEL_NAME = os.environ.get("MODEL_NAME", "openai/clip-vit-large-patch14")
 
+# Opcioni HF klasifikator (supervizirani model za Food-101). Preporučeni default: nateraw/food
+USE_HF_CLASSIFIER = os.environ.get("USE_HF_CLASSIFIER", "1") == "1"
+HF_FOOD_MODEL_NAME = os.environ.get("FOOD_CLASSIFIER_MODEL", "nateraw/food")
+
 # --- Helper Functions ---
 def select_device() -> str:
     """Odabire najbolji dostupni uređaj: CUDA > MPS (Apple) > CPU."""
@@ -103,6 +108,26 @@ def load_model():
         print(f"❌ Greška pri učitavanju CLIP modela: {e}")
         raise
 
+def load_hf_food_classifier(device: str):
+    """Pokušava učitati HF image-classification pipeline (npr. nateraw/food)."""
+    if not USE_HF_CLASSIFIER:
+        return None
+    try:
+        print(f"Loading HF Food Classifier: {HF_FOOD_MODEL_NAME} ...")
+        # Map device string -> pipeline device index
+        device_index = 0 if device in ("cuda", "mps") else -1
+        clf = hf_pipeline(
+            task="image-classification",
+            model=HF_FOOD_MODEL_NAME,
+            device=device_index,
+            top_k=5,
+        )
+        print("✅ HF Food Classifier učitan uspješno!")
+        return clf
+    except Exception as e:
+        print(f"⚠️ HF Food Classifier nije moguće učitati ('{HF_FOOD_MODEL_NAME}'): {e}. Nastavljam sa CLIP LITE.")
+        return None
+
 def is_image_file(file: UploadFile):
     """Provjerava da li je fajl podržani format slike."""
     return file.content_type in ["image/jpeg", "image/png", "image/jpg", "image/webp"]
@@ -342,6 +367,27 @@ def classify_image_with_clip(image: Image.Image, processor: CLIPProcessor, model
         "top5": list(zip(top_labels, top_probs))
     }
 
+def classify_image_with_hf(image: Image.Image, clf) -> Dict[str, Any]:
+    """Klasifikacija slike preko HF pipeline image-classification (top-5)."""
+    preds = clf(image)
+    # preds je lista dict-ova: {label, score}
+    if not preds:
+        return {
+            "primary_label": "Unknown",
+            "alternatives": [],
+            "confidence": 0.0,
+            "top5": []
+        }
+    top_labels = [p.get("label", "Unknown") for p in preds]
+    top_probs = [float(p.get("score", 0.0)) for p in preds]
+    primary_label = top_labels[0]
+    return {
+        "primary_label": primary_label,
+        "alternatives": top_labels[1:],
+        "confidence": top_probs[0] if top_probs else 0.0,
+        "top5": list(zip(top_labels, top_probs))
+    }
+
 def extract_clip_food_info(classification: Dict[str, Any]) -> Dict[str, Any]:
     """Formatira rezultat CLIP klasifikacije u zajedničku strukturu."""
     primary = classification["primary_label"]
@@ -363,16 +409,17 @@ def extract_clip_food_info(classification: Dict[str, Any]) -> Dict[str, Any]:
     }
 
 # --- Učitaj Model pri Pokretanju Aplikacije ---
-print("🚀 Pokrećem LITE Food Scanner API (CLIP)...")
+print("🚀 Pokrećem LITE Food Scanner API (CLIP + opcioni HF classifier)...")
 processor, model, device, dtype = load_model()
 CURRENT_DTYPE = dtype
 TEXT_LABELS = get_food101_labels()
 TEXT_FEATURES = build_text_cache(TEXT_LABELS, processor, model, device, dtype)
 warmup_model(processor, model, device, dtype)
+HF_CLASSIFIER = load_hf_food_classifier(device)
 
 # --- FastAPI Aplikacija ---
 app = FastAPI(
-    title="🍎 LITE Food Scanner API - Nutrition Edition (CLIP)",
+    title="🍎 LITE Food Scanner API - Nutrition Edition (HF+CLIP)",
     description="""
     **🏆 Lako i brzo prepoznavanje hrane + Nutrition Lookup (CPU-friendly)**
     
@@ -412,7 +459,7 @@ app = FastAPI(
     - 🤖 Inteligentna procjena za nepoznatu hranu
     - ✅ Production-ready i stabilan
     """,
-    version="9.0.0 - LITE (CLIP)"
+    version="9.1.0 - LITE (HF+CLIP)"
 )
 
 # Dodaj CORS middleware za web aplikacije
@@ -426,7 +473,7 @@ app.add_middleware(
 
 @app.post("/analyze", 
     summary="Analiziraj Food Sliku",
-    description="Upload-uj sliku da dobiješ food label + nutritivne podatke (CLIP LITE)",
+    description="Upload-uj sliku da dobiješ food label + nutritivne podatke (HF classifier ako je dostupan, inače CLIP LITE)",
     response_description="Rezultati food recognition i nutritivnih podataka"
 )
 async def analyze(file: UploadFile = File(...)):
@@ -464,10 +511,26 @@ async def analyze(file: UploadFile = File(...)):
         raise HTTPException(status_code=500, detail=f"Greška pri čitanju slike: {e}")
 
     try:
-        # Zero-shot klasifikacija sa CLIP-om
-        print("🔍 Analiziram sliku sa CLIP (zero-shot Food-101)...")
-        classification = classify_image_with_clip(image, processor, model, device)
-        food_info = extract_clip_food_info(classification)
+        # Ako postoji nadzirani HF classifier, koristi njega; inače CLIP zero-shot
+        if HF_CLASSIFIER is not None:
+            print("🔍 Analiziram sliku sa HF image-classification modelom ...")
+            classification = classify_image_with_hf(image, HF_CLASSIFIER)
+            # Prebaci u zajednički format
+            detailed = "Top-5: " + ", ".join([f"{l} ({p:.2f})" for l, p in classification["top5"]])
+            food_info = {
+                "primary_label": classification["primary_label"],
+                "alternative_labels": classification["alternatives"],
+                "detailed_analysis": detailed,
+                "food_items": f"1) {classification['primary_label']}",
+                "nutritional_context": "",
+                "ocr_text": "",
+                "has_food": True,
+                "confidence": classification["confidence"],
+            }
+        else:
+            print("🔍 Analiziram sliku sa CLIP (zero-shot Food-101)...")
+            classification = classify_image_with_clip(image, processor, model, device)
+            food_info = extract_clip_food_info(classification)
     except Exception as e:
         print(f"Greška tokom analize: {e}")
         raise HTTPException(status_code=500, detail=f"Greška tokom analize: {e}")
@@ -515,11 +578,15 @@ async def analyze(file: UploadFile = File(...)):
         },
         
         "model_info": {
-            "vision_model": MODEL_NAME,
+            "vision_model": HF_FOOD_MODEL_NAME if HF_CLASSIFIER is not None else MODEL_NAME,
             "nutrition_source": nutrition_data["source"],
-            "type": "CLIP Zero-shot Classifier + Nutrition Database",
+            "type": (
+                "HF Image Classification + Nutrition Database"
+                if HF_CLASSIFIER is not None else
+                "CLIP Zero-shot Classifier + Nutrition Database"
+            ),
             "capabilities": [
-                "Food Recognition (Food-101)",
+                "Food Recognition (supervised Food-101)" if HF_CLASSIFIER is not None else "Food Recognition (Food-101)",
                 "Nutrition Data Lookup"
             ]
         }
@@ -591,20 +658,20 @@ async def search_nutrition(food_name: str):
 def root():
     """Root endpoint sa API informacijama."""
     return {
-        "message": "🍎 LITE Food Scanner API v9.0 - CLIP + Nutrition Edition",
+        "message": "🍎 LITE Food Scanner API v9.1 - HF+CLIP + Nutrition Edition",
         "status": "🟢 Online",
         "tagline": "🏆 Najbolji Self-Hosted Food Recognition + Nutrition API",
         "model": {
-            "vision_model": MODEL_NAME,
+            "vision_model": HF_FOOD_MODEL_NAME if HF_CLASSIFIER is not None else MODEL_NAME,
             "nutrition_source": "Open Food Facts + AI Estimation",
-            "type": "CLIP Zero-shot Classifier + Nutrition Database",
-            "provider": "OpenAI CLIP + Open Food Facts",
-            "generation": "CLIP (ViT-L/14)",
+            "type": "HF Image Classification + Nutrition Database" if HF_CLASSIFIER is not None else "CLIP Zero-shot Classifier + Nutrition Database",
+            "provider": "Hugging Face + Open Food Facts" if HF_CLASSIFIER is not None else "OpenAI CLIP + Open Food Facts",
+            "generation": "ViT/ConvNeXt (supervised Food-101)" if HF_CLASSIFIER is not None else "CLIP (ViT-L/14)",
             "device": device.upper(),
             "rank": "🥇 LITE rješenje za Food Recognition"
         },
         "capabilities": {
-            "food_recognition": "✅ Food-101 Zero-shot (CLIP)",
+            "food_recognition": "✅ Food-101 (HF supervised)" if HF_CLASSIFIER is not None else "✅ Food-101 Zero-shot (CLIP)",
             "nutrition_data": "✅ Realne Nutritivne Vrijednosti",
             "nutrition_lookup": "✅ Manual Search po Imenu",
             "ingredient_analysis": "❌ (LITE)",
@@ -643,7 +710,7 @@ def root():
 )
 def health_check():
     """Health check endpoint za monitoring i load balancere."""
-    model_status = model is not None and processor is not None
+    model_status = (model is not None and processor is not None) or (HF_CLASSIFIER is not None)
     
     # Test nutrition API
     nutrition_api_status = "unknown"
@@ -656,12 +723,12 @@ def health_check():
     return {
         "status": "healthy" if model_status else "unhealthy",
         "model_loaded": model_status,
-        "vision_model": MODEL_NAME,
+        "vision_model": HF_FOOD_MODEL_NAME if HF_CLASSIFIER is not None else MODEL_NAME,
         "nutrition_api": nutrition_api_status,
-        "model_type": "CLIP Zero-shot Classifier + Nutrition Database",
+        "model_type": "HF Image Classification + Nutrition Database" if HF_CLASSIFIER is not None else "CLIP Zero-shot Classifier + Nutrition Database",
         "device": device,
         "device_available": torch.cuda.is_available() if device == "cuda" else True,
-        "version": "9.0.0 - LITE (CLIP)",
+        "version": "9.1.0 - LITE (HF+CLIP)",
         "timestamp": "2025-10-08",
         "ranking": "🥇 LITE Food Recognition + Nutrition Rješenje"
     }
@@ -675,7 +742,7 @@ def get_capabilities():
     return {
         "vision_model": MODEL_NAME,
         "nutrition_source": "Open Food Facts",
-        "generation": "CLIP (ViT-L/14) + Nutrition Database",
+            "generation": ("HF (supervised Food-101) + Nutrition Database" if HF_CLASSIFIER is not None else "CLIP (ViT-L/14) + Nutrition Database"),
         "release": "2024 (Stable)",
         "vision_tasks": {
             "food_recognition": {
@@ -725,8 +792,8 @@ def get_capabilities():
             "🔬 700,000+ proizvoda u bazi"
         ],
         "technical_specs": {
-            "parameters": "~427M",
-            "architecture": "CLIP (ViT-L/14)",
+            "parameters": "~86M-200M (ovisno o HF modelu)" if HF_CLASSIFIER is not None else "~427M",
+            "architecture": "HF Vision Classifier (npr. ViT-B/16)" if HF_CLASSIFIER is not None else "CLIP (ViT-L/14)",
             "training_data": "WIT + zero-shot na Food-101",
             "supported_formats": ["JPEG", "PNG", "WebP"],
             "max_resolution": "Podrška za visoke rezolucije",
@@ -741,12 +808,12 @@ def get_capabilities():
 # --- Pokreni API ---
 if __name__ == "__main__":
     print("=" * 80)
-    print("🍎 LITE FOOD SCANNER API v9.0 - NUTRITION EDITION (CLIP)")
+    print("🍎 LITE FOOD SCANNER API v9.1 - NUTRITION EDITION (HF+CLIP)")
     print("=" * 80)
-    print(f"🤖 Vision Model: {MODEL_NAME}")
+    print(f"🤖 Vision Model: {HF_FOOD_MODEL_NAME if HF_CLASSIFIER is not None else MODEL_NAME}")
     print(f"📊 Nutrition Source: Open Food Facts + AI Estimation")
-    print(f"🏢 Provider: OpenAI CLIP + Open Food Facts")
-    print(f"🔧 Type: CLIP Zero-shot Classifier + Nutrition Database")
+    print(f"🏢 Provider: {'Hugging Face + Open Food Facts' if HF_CLASSIFIER is not None else 'OpenAI CLIP + Open Food Facts'}")
+    print(f"🔧 Type: {'HF Image Classification + Nutrition Database' if HF_CLASSIFIER is not None else 'CLIP Zero-shot Classifier + Nutrition Database'}")
     print(f"💻 Device: {device.upper()}")
     print(f"🎯 Rank: LITE Food Recognition + Nutrition Rješenje")
     print(f"✨ Status: Production Ready - NUTRITION EDITION")