feature: access file system on mobile

app.py

@@ -1,19 +1,15 @@
-# app.py - FoodVision Streamlit Web Application
+# app.py - FoodVision Streamlit Web Application (Mobile-Optimized)
 # ============================================================
 # 
-# FEATURES:
-# ---------
-# ✅ Upload image (JPEG, PNG, WebP)
-# ✅ Take photo with camera (mobile/desktop)
-# ✅ Real-time prediction with confidence scores
-# ✅ Top-3 predictions display
-# ✅ Beautiful UI with image preview
-# ✅ Works on mobile & desktop
-#
-# DEPLOYMENT:
-# -----------
-# Local: streamlit run app.py
-# Cloud: Deploy to Streamlit Cloud, Hugging Face Spaces, or Railway
+# IMPROVEMENTS:
+# -------------
+# ✅ Mobile-friendly single-column layout
+# ✅ Auto-prediction on image upload (no button needed)
+# ✅ Simplified, responsive CSS
+# ✅ Better error handling
+# ✅ Loads model from Hugging Face Hub OR local file
+# ✅ Optimized for slow connections
+# ✅ Touch-friendly interface
 #
 # ============================================================
 
@@ -23,7 +19,6 @@ import torch.nn.functional as F
 from torchvision import transforms
 from PIL import Image
 import timm
-import numpy as np
 from pathlib import Path
 
 # ============================================================
@@ -31,75 +26,82 @@ from pathlib import Path
 # ============================================================
 
 st.set_page_config(
-    page_title="FoodVision AI - Food Classifier",
+    page_title="🍕 FoodVision AI",
     page_icon="🍕",
     layout="centered",
-    initial_sidebar_state="expanded"
+    initial_sidebar_state="collapsed"  # Better for mobile
 )
 
 # ============================================================
-# CUSTOM CSS FOR BEAUTIFUL UI
+# MINIMAL CSS (Mobile-First)
 # ============================================================
 
 st.markdown("""
 <style>
-    .main-header {
+    /* Remove extra padding on mobile */
+    .block-container {
+        padding-top: 2rem;
+        padding-bottom: 2rem;
+    }
+    
+    /* Cleaner header */
+    h1 {
         text-align: center;
         color: #FF6B6B;
-        font-size: 3rem;
-        font-weight: bold;
         margin-bottom: 0.5rem;
     }
-    .sub-header {
-        text-align: center;
-        color: #4ECDC4;
-        font-size: 1.2rem;
-        margin-bottom: 2rem;
-    }
-    .prediction-box {
+    
+    /* Result cards */
+    .prediction-card {
         background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
-        padding: 2rem;
-        border-radius: 15px;
+        padding: 1.5rem;
+        border-radius: 12px;
         color: white;
         text-align: center;
         margin: 1rem 0;
     }
-    .confidence-bar {
-        background-color: #f0f0f0;
-        border-radius: 10px;
-        height: 30px;
+    
+    .prediction-card h2 {
+        margin: 0;
+        font-size: 1.8rem;
+    }
+    
+    .prediction-card h3 {
+        margin: 0.5rem 0 0 0;
+        font-size: 1.2rem;
+        opacity: 0.9;
+    }
+    
+    /* Confidence bars */
+    .conf-bar {
+        background: #f0f0f0;
+        border-radius: 8px;
+        height: 36px;
         margin: 0.5rem 0;
         overflow: hidden;
+        position: relative;
     }
-    .confidence-fill {
+    
+    .conf-fill {
         height: 100%;
         background: linear-gradient(90deg, #4CAF50, #8BC34A);
         display: flex;
         align-items: center;
         justify-content: center;
         color: white;
-        font-weight: bold;
-        transition: width 0.5s ease;
-    }
-    .stButton>button {
-        width: 100%;
-        background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);
-        color: white;
-        font-size: 1.1rem;
-        padding: 0.75rem;
-        border-radius: 10px;
-        border: none;
-        font-weight: bold;
+        font-weight: 600;
+        font-size: 0.95rem;
     }
-    .stButton>button:hover {
-        background: linear-gradient(135deg, #764ba2 0%, #667eea 100%);
-        box-shadow: 0 4px 15px rgba(0,0,0,0.2);
+    
+    /* Info boxes */
+    .stAlert {
+        margin-top: 1rem;
     }
 </style>
 """, unsafe_allow_html=True)
 
 # ============================================================
-# FOOD CLASSES (101 categories)
+# FOOD CLASSES
 # ============================================================
 
 FOOD_CLASSES = [
@@ -126,65 +128,67 @@ FOOD_CLASSES = [
 ]
 
 # ============================================================
-# MODEL LOADING (WITH CACHING)
+# MODEL LOADING
 # ============================================================
 
 @st.cache_resource
-def load_model(model_path):
+def load_model():
     """
-    Loads the trained model with caching.
-    
-    Args:
-        model_path: Path to .pth checkpoint file
-    
-    Returns:
-        Loaded PyTorch model in eval mode
+    Loads model from local file or Hugging Face Hub.
+    Cached for performance across sessions.
     """
     try:
-        # Detect device
         device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
         
-        # Load checkpoint
-        checkpoint = torch.load(model_path, map_location=device)
+        # Try loading from local file first (for HF Spaces)
+        local_path = Path("model1_best.pth")
+        
+        if local_path.exists():
+            checkpoint = torch.load(local_path, map_location=device)
+        else:
+            # Fallback: try to download from HF Hub
+            try:
+                from huggingface_hub import hf_hub_download
+                model_path = hf_hub_download(
+                    repo_id="doozer21/FoodVision",
+                    filename="model1_best.pth"
+                )
+                checkpoint = torch.load(model_path, map_location=device)
+            except Exception as e:
+                st.error("❌ Could not load model from local file or Hugging Face Hub")
+                st.info("Make sure model1_best.pth is in your Space's repository")
+                return None, None, None
         
-        # Get model config from checkpoint
+        # Get config
         model_config = checkpoint.get('model_config', {
             'model_id': 'convnextv2_base.fcmae_ft_in22k_in1k_384'
         })
         
-        # Create model architecture
+        # Create and load model
         model = timm.create_model(
             model_config['model_id'],
             pretrained=False,
             num_classes=101
         )
         
-        # Load trained weights
         model.load_state_dict(checkpoint['model_state_dict'])
         model.to(device)
         model.eval()
         
-        return model, device, checkpoint.get('best_val_acc', 0)
+        accuracy = checkpoint.get('best_val_acc', 0)
+        
+        return model, device, accuracy
         
     except Exception as e:
         st.error(f"❌ Error loading model: {str(e)}")
-        st.info("💡 Make sure 'model1_best.pth' is in the same directory as app.py")
-        return None, None, 0
+        return None, None, None
 
 # ============================================================
 # IMAGE PREPROCESSING
 # ============================================================
 
 def preprocess_image(image):
-    """
-    Preprocesses image for model input.
-    
-    Args:
-        image: PIL Image
-    
-    Returns:
-        Preprocessed tensor ready for model
-    """
+    """Preprocess image for model input."""
     transform = transforms.Compose([
         transforms.Resize(256),
         transforms.CenterCrop(224),
@@ -195,56 +199,29 @@ def preprocess_image(image):
         )
     ])
     
-    # Convert to RGB (handle PNG with alpha, grayscale, etc.)
     image = image.convert('RGB')
-    
-    # Apply transforms
-    img_tensor = transform(image).unsqueeze(0)  # Add batch dimension
-    
-    return img_tensor
+    return transform(image).unsqueeze(0)
 
 # ============================================================
-# PREDICTION FUNCTION
+# PREDICTION
 # ============================================================
 
 def predict(model, image_tensor, device, top_k=3):
-    """
-    Makes prediction on preprocessed image.
-    
-    Args:
-        model: Trained PyTorch model
-        image_tensor: Preprocessed image tensor
-        device: torch device
-        top_k: Number of top predictions to return
-    
-    Returns:
-        List of (class_name, confidence) tuples
-    """
+    """Make prediction on image."""
     with torch.no_grad():
-        # Move to device
         image_tensor = image_tensor.to(device)
-        
-        # Forward pass
         outputs = model(image_tensor)
-        
-        # Get probabilities
         probabilities = F.softmax(outputs, dim=1)
         
-        # Get top-k predictions
         top_probs, top_indices = torch.topk(probabilities, top_k)
-        
-        # Convert to Python lists
         top_probs = top_probs.cpu().numpy()[0]
         top_indices = top_indices.cpu().numpy()[0]
         
-        # Create results
         results = []
         for prob, idx in zip(top_probs, top_indices):
-            class_name = FOOD_CLASSES[idx]
-            # Format class name (replace underscores, title case)
-            formatted_name = class_name.replace('_', ' ').title()
+            class_name = FOOD_CLASSES[idx].replace('_', ' ').title()
             confidence = float(prob) * 100
-            results.append((formatted_name, confidence))
+            results.append((class_name, confidence))
         
         return results
 
@@ -254,159 +231,144 @@ def predict(model, image_tensor, device, top_k=3):
 
 def main():
     # Header
-    st.markdown('<h1 class="main-header">🍕 FoodVision AI</h1>', unsafe_allow_html=True)
-    st.markdown('<p class="sub-header">Identify 101 food dishes with AI-powered computer vision</p>', unsafe_allow_html=True)
-    
-    # Sidebar
-    with st.sidebar:
-        st.header("📊 Model Information")
-        st.write("**Architecture:** ConvNeXt V2 Base")
-        st.write("**Training:** Food-101 Dataset")
-        st.write("**Classes:** 101 food categories")
-        
-        st.markdown("---")
-        
-        st.header("🎯 How to Use")
-        st.write("1. Upload a food image or take a photo")
-        st.write("2. Wait for AI analysis")
-        st.write("3. View top-3 predictions")
-        
-        st.markdown("---")
-        
-        st.header("🔗 Resources")
-        st.markdown("[Food-101 Dataset](https://data.vision.ee.ethz.ch/cvl/datasets_extra/food-101/)")
-        st.markdown("[ConvNeXt V2 Paper](https://arxiv.org/abs/2301.00808)")
-    
-    # Load model
-    model_path = "model1_best.pth"
-    
-    if not Path(model_path).exists():
-        st.error(f"❌ Model file '{model_path}' not found!")
-        st.info("💡 Please place your trained model file in the same directory as app.py")
-        st.stop()
+    st.title("🍕 FoodVision AI")
+    st.markdown("**Identify 101 food dishes instantly**")
     
+    # Load model with status
     with st.spinner("🔄 Loading AI model..."):
-        model, device, accuracy = load_model(model_path)
+        model, device, accuracy = load_model()
     
     if model is None:
         st.stop()
     
-    st.success(f"✅ Model loaded! Accuracy: {accuracy:.2f}%")
-    
-    # Update sidebar with actual accuracy
-    with st.sidebar:
+    # Show model info in expander (cleaner for mobile)
+    with st.expander("ℹ️ Model Info"):
+        st.write(f"**Architecture:** ConvNeXt V2 Base")
         st.write(f"**Accuracy:** {accuracy:.2f}%")
+        st.write(f"**Device:** {'GPU' if device.type == 'cuda' else 'CPU'}")
+        st.write(f"**Classes:** 101 food categories")
     
-    # Main content area
     st.markdown("---")
     
-    # Two columns for input methods
-    col1, col2 = st.columns(2)
+    # Single-column layout (mobile-friendly)
+    st.subheader("📸 Upload or Take a Photo")
     
-    with col1:
-        st.subheader("📁 Upload Image")
-        uploaded_file = st.file_uploader(
-            "Choose a food image",
-            type=['jpg', 'jpeg', 'png', 'webp'],
-            help="Supported formats: JPG, JPEG, PNG, WebP"
-        )
+    # File uploader
+    uploaded_file = st.file_uploader(
+        "Choose a food image",
+        type=['jpg', 'jpeg', 'png', 'webp'],
+        label_visibility="collapsed"
+    )
     
-    with col2:
-        st.subheader("📸 Take Photo")
-        camera_photo = st.camera_input(
-            "Take a picture",
-            help="Works on mobile and desktop with camera"
-        )
+    # Camera input (below uploader)
+    st.markdown("**Or use your camera:**")
+    camera_photo = st.camera_input(
+        "Take a picture",
+        label_visibility="collapsed"
+    )
     
-    # Process image (either uploaded or from camera)
+    # Determine which image to use
     image_source = None
+    source_name = ""
     
-    if uploaded_file is not None:
-        image_source = uploaded_file
-        source_type = "uploaded"
-    elif camera_photo is not None:
+    if camera_photo is not None:
         image_source = camera_photo
-        source_type = "camera"
+        source_name = "camera"
+    elif uploaded_file is not None:
+        image_source = uploaded_file
+        source_name = "upload"
     
-    # If we have an image, process it
+    # Process image automatically (no button needed!)
     if image_source is not None:
         try:
             # Load image
             image = Image.open(image_source)
             
-            # Display image
+            # Show image preview
+            st.image(image, caption=f"Image from {source_name}", use_column_width=True)
+            
+            # Auto-predict with spinner
+            with st.spinner("🧠 Analyzing your food..."):
+                img_tensor = preprocess_image(image)
+                predictions = predict(model, img_tensor, device, top_k=3)
+            
             st.markdown("---")
-            st.subheader("📷 Your Image")
-            st.image(image, use_container_width=True, caption=f"Image from {source_type}")
             
-            # Predict button
-            if st.button("🔮 Analyze Food", use_container_width=True):
-                with st.spinner("🧠 AI is analyzing your food..."):
-                    # Preprocess
-                    img_tensor = preprocess_image(image)
-                    
-                    # Predict
-                    predictions = predict(model, img_tensor, device, top_k=3)
-                
-                # Display results
-                st.markdown("---")
-                st.subheader("🎯 Prediction Results")
-                
-                # Top prediction (large display)
-                top_food, top_conf = predictions[0]
+            # Display top prediction prominently
+            top_food, top_conf = predictions[0]
+            
+            st.markdown(f"""
+            <div class="prediction-card">
+                <h2>🏆 {top_food}</h2>
+                <h3>{top_conf:.1f}% Confidence</h3>
+            </div>
+            """, unsafe_allow_html=True)
+            
+            # Show all top-3 predictions
+            st.markdown("### 📊 Top 3 Predictions")
+            
+            for i, (food, conf) in enumerate(predictions, 1):
+                emoji = "🥇" if i == 1 else "🥈" if i == 2 else "🥉"
                 
+                st.markdown(f"**{emoji} {food}**")
                 st.markdown(f"""
-                <div class="prediction-box">
-                    <h2>🏆 {top_food}</h2>
-                    <h3>{top_conf:.1f}% Confidence</h3>
+                <div class="conf-bar">
+                    <div class="conf-fill" style="width: {conf}%">
+                        {conf:.1f}%
+                    </div>
                 </div>
                 """, unsafe_allow_html=True)
-                
-                # Top-3 predictions with bars
-                st.markdown("### 📊 Top 3 Predictions")
-                
-                for i, (food, conf) in enumerate(predictions, 1):
-                    # Emoji for rank
-                    emoji = "🥇" if i == 1 else "🥈" if i == 2 else "🥉"
-                    
-                    # Display with confidence bar
-                    st.write(f"**{emoji} {food}**")
-                    st.markdown(f"""
-                    <div class="confidence-bar">
-                        <div class="confidence-fill" style="width: {conf}%">
-                            {conf:.1f}%
-                        </div>
-                    </div>
-                    """, unsafe_allow_html=True)
-                
-                # Additional info
-                st.info(f"💡 **Tip:** The model is {top_conf:.1f}% confident this is {top_food.lower()}!")
-                
-                # Fun facts (optional)
-                if top_conf > 90:
-                    st.success("🎉 Very high confidence! The model is very sure about this prediction.")
-                elif top_conf > 70:
-                    st.success("👍 Good confidence! This looks like a solid prediction.")
-                else:
-                    st.warning("🤔 Moderate confidence. The food might be ambiguous or partially visible.")
+            
+            # Feedback based on confidence
+            st.markdown("---")
+            if top_conf > 90:
+                st.success("🎉 **Very confident!** The model is very sure about this prediction.")
+            elif top_conf > 70:
+                st.success("👍 **Good confidence!** This looks like a solid prediction.")
+            elif top_conf > 50:
+                st.warning("🤔 **Moderate confidence.** The food might be ambiguous or partially visible.")
+            else:
+                st.warning("😕 **Low confidence.** Try a clearer photo with better lighting.")
         
         except Exception as e:
-            st.error(f"❌ Error processing image: {str(e)}")
-            st.info("💡 Try a different image or check if the file is corrupted")
+            st.error(f"❌ Error: {str(e)}")
+            st.info("Try a different image or check if the file is corrupted")
     
     else:
-        # No image yet - show example
-        st.info("👆 Upload an image or take a photo to get started!")
+        # Instructions
+        st.info("👆 Upload a food image or take a photo to get started!")
+        
+        with st.expander("💡 Tips for Best Results"):
+            st.markdown("""
+            - Use clear, well-lit photos
+            - Make sure food is the main subject
+            - Avoid heavily filtered images
+            - Try different angles if confidence is low
+            - Works best with common dishes
+            """)
         
-        st.markdown("---")
-        st.subheader("💡 Tips for Best Results")
-        st.write("• Use clear, well-lit photos")
-        st.write("• Make sure the food is the main subject")
-        st.write("• Avoid heavily filtered or edited images")
-        st.write("• Try different angles if confidence is low")
+        with st.expander("🍽️ What can it recognize?"):
+            st.markdown("""
+            The model can identify **101 popular dishes** including:
+            - 🍕 Pizza, Pasta, Burgers
+            - 🍣 Sushi, Ramen, Pad Thai
+            - 🥗 Salads, Sandwiches
+            - 🍰 Desserts (cakes, ice cream, etc.)
+            - 🍳 Breakfast foods
+            - And many more!
+            """)
+    
+    # Footer
+    st.markdown("---")
+    st.markdown(
+        "<div style='text-align: center; color: #666; font-size: 0.9rem;'>"
+        "Built with Streamlit • ConvNeXt V2 • Food-101 Dataset"
+        "</div>",
+        unsafe_allow_html=True
+    )
 
 # ============================================================
-# RUN APP
+# RUN
 # ============================================================
 
 if __name__ == "__main__":