Update app.py

app.py

@@ -4,7 +4,30 @@ import numpy as np
 from PIL import Image
 import json
 import os
-from huggingface_hub import hf_hub_download
+
+# Define custom metrics (MUST be defined before loading model)
+def top_3_accuracy(y_true, y_pred):
+    return tf.keras.metrics.top_k_categorical_accuracy(y_true, y_pred, k=3)
+
+def top_5_accuracy(y_true, y_pred):
+    return tf.keras.metrics.top_k_categorical_accuracy(y_true, y_pred, k=5)
+
+# Register custom functions with Keras
+@tf.keras.utils.register_keras_serializable()
+def top_3_accuracy_serializable(y_true, y_pred):
+    return tf.keras.metrics.top_k_categorical_accuracy(y_true, y_pred, k=3)
+
+@tf.keras.utils.register_keras_serializable()
+def top_5_accuracy_serializable(y_true, y_pred):
+    return tf.keras.metrics.top_k_categorical_accuracy(y_true, y_pred, k=5)
+
+# Custom objects dictionary
+custom_objects = {
+    'top_3_accuracy': top_3_accuracy,
+    'top_5_accuracy': top_5_accuracy,
+    'top_3_accuracy_serializable': top_3_accuracy_serializable,
+    'top_5_accuracy_serializable': top_5_accuracy_serializable
+}
 
 # Page configuration
 st.set_page_config(
@@ -71,66 +94,72 @@ st.markdown('<div class="pakistan-flag">🇵🇰 Pakistan 🇵🇰</div>', unsaf
 MODEL_PATH = "dish_classifier_final.keras"
 CLASSES_PATH = "class_names.json"
 
-# Try to download model from Hugging Face cache if not exists
 @st.cache_resource
 def load_model():
-    """Load the trained model from local or Hugging Face cache"""
+    """Load the trained model with custom objects"""
     
-    # First check if model exists locally
-    if os.path.exists(MODEL_PATH):
-        try:
-            model = tf.keras.models.load_model(MODEL_PATH)
-            st.success("✅ Model loaded successfully from local file!")
-            return model
-        except Exception as e:
-            st.error(f"Error loading local model: {e}")
+    # Check if model exists
+    if not os.path.exists(MODEL_PATH):
+        st.error(f"""
+        ❌ Model file not found: `{MODEL_PATH}`
+        
+        **Please upload the model file:**
+        1. Click on **"Files"** tab above
+        2. Click **"Add file"** → **"Upload file"**
+        3. Upload `dish_classifier_final.keras`
+        4. Upload `class_names.json`
+        5. Refresh this page
+        
+        **Note:** The model file should be around 80-100 MB.
+        """)
+        return None
     
-    # If not found locally, try to find in Hugging Face cache
     try:
-        # Try to download from Hugging Face (if uploaded)
-        model_path = hf_hub_download(
-            repo_id=st.secrets.get("HF_REPO_ID", "your-username/your-space"),
-            filename="dish_classifier_final.keras",
-            repo_type="space"
+        # Load model with custom objects
+        model = tf.keras.models.load_model(
+            MODEL_PATH,
+            custom_objects=custom_objects,
+            compile=False
         )
-        model = tf.keras.models.load_model(model_path)
-        st.success("✅ Model loaded from Hugging Face cache!")
-        return model
-    except:
-        st.error("""
-        ⚠️ Model file not found!
         
-        **Please upload your model file:**
-        1. Click on "Files" tab above
-        2. Upload `dish_classifier_final.keras`
-        3. Upload `class_names.json`
-        4. Refresh this page
-        """)
+        # Recompile the model with the same metrics
+        model.compile(
+            optimizer='adam',
+            loss='categorical_crossentropy',
+            metrics=['accuracy', top_3_accuracy, top_5_accuracy]
+        )
+        
+        st.success("✅ Model loaded successfully!")
+        return model
+    except Exception as e:
+        st.error(f"Error loading model: {str(e)}")
+        st.info("Make sure the model file is not corrupted and was trained with TensorFlow 2.x")
         return None
 
 @st.cache_data
 def load_class_names():
     """Load class names from JSON file"""
+    if not os.path.exists(CLASSES_PATH):
+        st.error(f"❌ Class names file not found: `{CLASSES_PATH}`")
+        st.info("Please upload `class_names.json` file")
+        return None
+    
     try:
-        if os.path.exists(CLASSES_PATH):
-            with open(CLASSES_PATH, 'r') as f:
-                class_names = json.load(f)
-            st.success(f"✅ Loaded {len(class_names)} Pakistan food classes!")
-            return class_names
-        else:
-            st.error("class_names.json not found! Please upload the file.")
-            return None
+        with open(CLASSES_PATH, 'r') as f:
+            class_names = json.load(f)
+        st.success(f"✅ Loaded {len(class_names)} Pakistani food classes!")
+        return class_names
     except Exception as e:
         st.error(f"Error loading class names: {e}")
         return None
 
 def preprocess_image(image):
     """Preprocess image for model prediction"""
-    # Convert to RGB (remove alpha channel if present)
+    # Convert to RGB if needed
     if image.mode != 'RGB':
         image = image.convert('RGB')
     
-    # Resize to 224x224 (model input size)
+    # Resize to 224x224
     image = image.resize((224, 224))
     
     # Convert to array
@@ -207,7 +236,7 @@ def main():
         st.markdown("---")
         st.markdown("### 📊 Model Performance")
         
-        # Add a simple gauge for accuracy
+        # Add simple gauges for accuracy
         st.markdown("**Exact Match Accuracy**")
         st.progress(0.56)
         st.caption("56.25%")
@@ -297,14 +326,20 @@ def main():
                     
                     # Add confidence visualization
                     st.markdown("### 📊 Confidence Level")
-                    confidence_color = "🟢" if top_5_probs[0] > 70 else "🟡" if top_5_probs[0] > 50 else "🟠"
-                    st.markdown(f"{confidence_color} Model is {top_5_probs[0]:.1f}% confident this is **{top_5_labels[0]}**")
+                    if top_5_probs[0] > 70:
+                        confidence_color = "🟢 High Confidence"
+                    elif top_5_probs[0] > 50:
+                        confidence_color = "🟡 Medium Confidence"
+                    else:
+                        confidence_color = "🟠 Low Confidence"
                     
-                    # Add suggestion
-                    if top_5_probs[0] < 50:
-                        st.info("💡 Tip: Try uploading a clearer image of the dish alone for better accuracy")
+                    st.markdown(f"{confidence_color} - {top_5_probs[0]:.1f}% confident this is **{top_5_labels[0]}**")
                     
-    # Add some Pakistan food facts
+                    # Add suggestion for low confidence
+                    if top_5_probs[0] < 50:
+                        st.info("💡 **Tip:** Try uploading a clearer image of just the dish for better accuracy")
+    
+    # Add information when no image is uploaded
     if not uploaded_file:
         st.markdown("---")
         st.markdown("### 🕌 About This Model")