Add ConvNeXt Large model and app files
bea5a24 over 1 year ago
28.7 kB
import streamlit as st
from PIL import Image
import torch
from torchvision import models
import torch.nn as nn
import torchvision.transforms as transforms
import io
import os
import openai
import json
import timm  # For ConvNeXt Large model

# --- Page Configuration ---
st.set_page_config(page_title="EatSmart Pro", page_icon="🍽️", layout="wide")

# Mobile menu indicator
st.markdown("""
<div style="display: block; background: linear-gradient(90deg, #667eea 0%, #764ba2 100%); 
           padding: 8px 15px; border-radius: 8px; margin-bottom: 15px; text-align: center;">
    <p style="color: white; margin: 0; font-size: 14px;">
        📱 <strong>Mobile Users:</strong> Click the <strong>></strong> arrow (top-left) to open preferences menu
    </p>
</div>
""", unsafe_allow_html=True)

# ==============================================================================
# The correct, full list of 101 class names from your training script.
# ==============================================================================
CLASS_NAMES = [
    'apple_pie', 'baby_back_ribs', 'baklava', 'beef_carpaccio', 'beef_tartare',
    'beet_salad', 'beignets', 'bibimbap', 'bread_pudding', 'breakfast_burrito',
    'bruschetta', 'caesar_salad', 'cannoli', 'caprese_salad', 'carrot_cake',
    'ceviche', 'cheesecake', 'cheese_plate', 'chicken_curry', 'chicken_quesadilla',
    'chicken_wings', 'chocolate_cake', 'chocolate_mousse', 'churros', 'clam_chowder',
    'club_sandwich', 'crab_cakes', 'creme_brulee', 'croque_madame', 'cup_cakes',
    'deviled_eggs', 'donuts', 'dumplings', 'edamame', 'eggs_benedict',
    'escargots', 'falafel', 'filet_mignon', 'fish_and_chips', 'foie_gras',
    'french_fries', 'french_onion_soup', 'french_toast', 'fried_calamari', 'fried_rice',
    'frozen_yogurt', 'garlic_bread', 'gnocchi', 'greek_salad', 'grilled_cheese_sandwich',
    'grilled_salmon', 'guacamole', 'gyoza', 'hamburger', 'hot_and_sour_soup',
    'hot_dog', 'huevos_rancheros', 'hummus', 'ice_cream', 'lasagna',
    'lobster_bisque', 'lobster_roll_sandwich', 'macaroni_and_cheese', 'macarons', 'miso_soup',
    'mussels', 'nachos', 'omelette', 'onion_rings', 'oysters',
    'pad_thai', 'paella', 'pancakes', 'panna_cotta', 'peking_duck',
    'pho', 'pizza', 'pork_chop', 'poutine', 'prime_rib',
    'pulled_pork_sandwich', 'ramen', 'ravioli', 'red_velvet_cake', 'risotto',
    'samosa', 'sashimi', 'scallops', 'seaweed_salad', 'shrimp_and_grits',
    'spaghetti_bolognese', 'spaghetti_carbonara', 'spring_rolls', 'steak', 'strawberry_shortcake',
    'sushi', 'tacos', 'takoyaki', 'tiramisu', 'tuna_tartare',
    'waffles'
]

# ==============================================================================
# ALL HELPER FUNCTIONS ARE NOW INSIDE A SINGLE APP.PY
# ==============================================================================

def get_convnext_model(num_classes):
    """
    Creates the ConvNeXt Large model architecture,
    matching the training script for maximum accuracy.
    """
    print(f"🚀 Loading ConvNeXt Large model for inference...")
    print(f"📊 Model: ConvNeXt Large (197M parameters)")
    
    # Create ConvNeXt Large model (same as training script)
    model = timm.create_model('convnext_large.fb_in22k_ft_in1k', pretrained=True, num_classes=num_classes)
    
    # Model statistics for user feedback
    total_params = sum(p.numel() for p in model.parameters())
    print(f"✅ Model architecture loaded:")
    print(f"   Total parameters: {total_params:,}")
    print(f"   Model size: ~{total_params * 4 / 1024**2:.1f} MB")
    
    return model

def get_efficientnet_model(num_classes):
    """
    Creates the old EfficientNet-V2-S model architecture for fallback.
    """
    print(f"⚠️ Loading EfficientNet-V2-S model (fallback)...")
    model = models.efficientnet_v2_s(weights=None)
    num_features = model.classifier[1].in_features
    model.classifier = nn.Sequential(
        nn.Dropout(p=0.2),
        nn.Linear(num_features, 256),
        nn.ReLU(),
        nn.Dropout(p=0.1),
        nn.Linear(256, num_classes)
    )
    return model

def load_json_data(path):
    if not os.path.exists(path): return {}
    try:
        with open(path, 'r') as f: return json.load(f)
    except (FileNotFoundError, json.JSONDecodeError): return {}

def get_health_info(food_name, health_data):
    """Enhanced health information display with comprehensive nutritional data"""
    food_name_key = food_name.replace('_', ' ').title()
    
    # Try to get from the nested nutrition_info structure first
    nutrition_info = health_data.get("nutrition_info", {})
    info = nutrition_info.get(food_name_key) or nutrition_info.get(food_name.lower())
    
    # Try direct lookup if nested lookup fails
    if not info:
        info = health_data.get(food_name_key)
    
    if not info:
        return "<p>No specific health information available for this dish.</p>"
    
    # Get health score
    health_scores = health_data.get("health_scores", {})
    health_score = health_scores.get(food_name.lower(), {})
    score = health_score.get("score", 75)
    score_explanation = health_score.get("explanation", "Good")
    
    # Determine score color
    if score >= 80:
        score_color = "#28a745"  # Green
        score_bg = "#d4edda"
    elif score >= 60:
        score_color = "#ffc107"  # Yellow
        score_bg = "#fff3cd"
    else:
        score_color = "#dc3545"  # Red
        score_bg = "#f8d7da"
    
    # Build nutritional information
    nutrition_html = f"""
    <div style="background-color: #f8f9fa; padding: 15px; border-radius: 8px; margin: 10px 0;">
        <h4 style="color: #495057; margin-top: 0;">📊 Nutritional Information</h4>
        <div style="display: grid; grid-template-columns: repeat(4, 1fr); gap: 15px; text-align: center;">
            <div>
                <div style="font-size: 14px; color: #6c757d;">Calories</div>
                <div style="font-size: 24px; font-weight: bold; color: #495057;">{info.get("calories", "N/A")}</div>
                <div style="font-size: 12px; color: #6c757d;">kcal</div>
            </div>
            <div>
                <div style="font-size: 14px; color: #6c757d;">Protein</div>
                <div style="font-size: 24px; font-weight: bold; color: #495057;">{info.get("protein", "N/A")}</div>
                <div style="font-size: 12px; color: #6c757d;">g</div>
            </div>
            <div>
                <div style="font-size: 14px; color: #6c757d;">Carbs</div>
                <div style="font-size: 24px; font-weight: bold; color: #495057;">{info.get("carbs", "N/A")}</div>
                <div style="font-size: 12px; color: #6c757d;">g</div>
            </div>
            <div>
                <div style="font-size: 14px; color: #6c757d;">Fat</div>
                <div style="font-size: 24px; font-weight: bold; color: #495057;">{info.get("fat", "N/A")}</div>
                <div style="font-size: 12px; color: #6c757d;">g</div>
            </div>
        </div>
    </div>
    """
    
    # Health Score
    score_html = f"""
    <div style="background-color: {score_bg}; border: 1px solid {score_color}; border-radius: 8px; padding: 15px; margin: 10px 0;">
        <h4 style="color: {score_color}; margin-top: 0;">🏥 Health Assessment</h4>
        <div style="background-color: {score_color}; color: white; padding: 10px; border-radius: 5px; text-align: center; font-weight: bold;">
            Health Score: {score}/100 ({score_explanation})
        </div>
    </div>
    """
    
    # Health Benefits - Display using Streamlit components instead of HTML
    benefits_section = ""
    if info.get("benefits"):
        # We'll handle benefits separately using st.markdown with proper formatting
        benefits_section = "BENEFITS_SECTION"  # Placeholder to indicate benefits exist
    
    return nutrition_html + score_html + benefits_section

def get_allergen_info(food_name, allergen_data):
    """Enhanced allergen information using sidebar preferences"""
    food_name_key = food_name.replace('_', ' ').title()
    allergens = allergen_data.get(food_name_key, [])
    
    # Display allergen information for the detected food
    if allergens:
        # Check for user-specified allergen matches
        user_allergen_matches = [a for a in allergens if a in st.session_state.user_allergens]
        
        if user_allergen_matches:
            # High priority alert for user's allergens
            st.error(f"🚨 **CRITICAL ALLERGEN ALERT**: This dish contains **{', '.join(user_allergen_matches)}** which you've marked as allergens to avoid!")
        
        # Display allergens in a more user-friendly way
        st.markdown("### ⚠️ Allergens Detected in This Food")
        
        # Create columns for allergen badges
        cols = st.columns(min(len(allergens), 4))
        for i, allergen in enumerate(allergens):
            with cols[i % len(cols)]:
                if allergen in st.session_state.user_allergens:
                    # User's marked allergens - show as error
                    st.error(f"🚨 {allergen}")
                else:
                    # Other allergens - show as info
                    st.info(f"ℹ️ {allergen}")
        
        # Add explanation
        if user_allergen_matches:
            st.markdown("---")
            st.markdown("🔴 **Red alerts** are for allergens you've marked in your preferences")
        else:
            st.markdown("---")
            st.markdown("💙 **Blue badges** show allergens present in this food")
            
    else:
        st.success("✅ No common allergens typically found in this dish.")

def get_trans_fat_analysis(food_name, health_data):
    """Enhanced trans fat analysis with user preferences"""
    food_name_lower = food_name.lower().replace('_', ' ')
    
    # Get trans fat ingredients from health data
    trans_fat_ingredients = health_data.get("trans_fat_ingredients", [
        "hydrogenated oil", "partially hydrogenated oil", "margarine", "shortening"
    ])
    
    # Foods that commonly contain trans fats
    high_trans_fat_foods = [
        "donuts", "french_fries", "fried", "margarine", "shortening", 
        "processed", "packaged", "fast food", "baked goods", "cookies", "crackers"
    ]
    
    # Check if food likely contains trans fats
    likely_trans_fat = any(ingredient in food_name_lower for ingredient in high_trans_fat_foods)
    
    if likely_trans_fat:
        alert_level = "HIGH RISK" if st.session_state.avoid_trans_fat else "POTENTIAL RISK"
        if st.session_state.avoid_trans_fat:
            st.error(f"🚨 **TRANS FAT ALERT**: You've enabled trans fat warnings and this food may contain trans fats!")
        
        trans_fat_html = f"""
        <div style="background: linear-gradient(135deg, #f8d7da 0%, #f5c6cb 100%); border: 2px solid #dc3545; border-radius: 12px; padding: 20px; margin: 15px 0; box-shadow: 0 4px 6px rgba(0,0,0,0.1);">
            <h4 style="color: #721c24; margin-top: 0; text-align: center;">🧪 Trans Fat Analysis - {alert_level}</h4>
            <div style="color: #721c24; text-align: center;">
                <p><strong>⚠️ This food may contain trans fats from:</strong></p>
                <div style="display: flex; flex-wrap: wrap; justify-content: center; gap: 10px; margin: 15px 0;">
                    {"".join([f'<span style="background-color: #dc3545; color: white; padding: 5px 10px; border-radius: 15px; font-size: 0.9em;">{ingredient.title()}</span>' for ingredient in trans_fat_ingredients])}
                </div>
                <p style="font-weight: bold; margin-top: 15px;">💡 <em>Recommendation: Check ingredient labels and consider healthier alternatives</em></p>
            </div>
        </div>
        """
    else:
        trans_fat_html = f"""
        <div style="background: linear-gradient(135deg, #d4edda 0%, #c3e6cb 100%); border: 2px solid #28a745; border-radius: 12px; padding: 20px; margin: 15px 0; box-shadow: 0 4px 6px rgba(0,0,0,0.1);">
            <h4 style="color: #155724; margin-top: 0; text-align: center;">🧪 Trans Fat Analysis - LOW RISK</h4>
            <div style="color: #155724; text-align: center;">
                <p><strong>✅ Great Choice!</strong></p>
                <p>This food typically contains minimal or no artificial trans fats.</p>
                <p style="font-style: italic; margin-top: 10px;">Keep enjoying healthy foods like this! 🌟</p>
            </div>
        </div>
        """
    
    return trans_fat_html

def generate_recipe(food_name):
    """Enhanced recipe generation based on detected ingredients and user preferences"""
    try:
        api_key = st.secrets.get("OPENAI_API_KEY")
        if not api_key: 
            return "Error: OPENAI_API_KEY not found."
        
        openai.api_key = api_key
        
        # Build prompt based on user preferences
        dietary_restrictions = ""
        if st.session_state.dietary_preferences:
            dietary_restrictions = f"Make it {', '.join(st.session_state.dietary_preferences).lower()}. "
        
        allergen_restrictions = ""
        if st.session_state.user_allergens:
            allergen_restrictions = f"Avoid using {', '.join(st.session_state.user_allergens).lower()}. "
        
        trans_fat_note = ""
        if st.session_state.avoid_trans_fat:
            trans_fat_note = "Use healthy oils and avoid trans fats. "
        
        prompt = f"""Create a recipe for {food_name}. {dietary_restrictions}{allergen_restrictions}{trans_fat_note}
        
        Format the response with these exact headings:
        Ingredients:
        Instructions:
        Chef's Tips:
        
        Make it practical for home cooking and include nutritional benefits."""
        
        response = openai.chat.completions.create(
            model="gpt-3.5-turbo",
            messages=[
                {"role": "system", "content": "You are a professional chef who creates healthy, personalized recipes based on dietary preferences and restrictions."},
                {"role": "user", "content": prompt}
            ],
            temperature=0.7, 
            max_tokens=600
        )
        
        recipe_text = response.choices[0].message.content
        return recipe_text.replace("**Ingredients:**", "Ingredients:").replace("**Instructions:**", "Instructions:").replace("**Chef's Tips:**", "Chef's Tips:")
    except Exception as e: 
        return f"The AI Chef is busy right now. Error: {str(e)}"

@st.cache_resource
def load_model_resources():
    try:
        num_classes = len(CLASS_NAMES)
        
        # Look for the new ConvNeXt Large model first, fallback to old models
        convnext_model_path = 'models/food_classifier_convnext_large_cpu_full.pth'
        efficientnet_model_path = 'models/food101_efficientnet_best.pth'
        old_model_path = 'models/final_model.pth'
        
        if os.path.exists(convnext_model_path):
            model_path = convnext_model_path
            model = get_convnext_model(num_classes=num_classes)
            print(f"🎯 Using NEW ConvNeXt Large model: {model_path}")
        elif os.path.exists(efficientnet_model_path):
            model_path = efficientnet_model_path
            model = get_efficientnet_model(num_classes=num_classes)
            print(f"⚠️ Using EfficientNet model: {model_path}")
        elif os.path.exists(old_model_path):
            model_path = old_model_path
            model = get_efficientnet_model(num_classes=num_classes)
            print(f"⚠️ Using old fallback model: {model_path}")
        else:
            st.error(f"FATAL: No model file found. Looking for:\n- {convnext_model_path}\n- {efficientnet_model_path}\n- {old_model_path}")
            return None, None, None, None
        
        device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
        checkpoint = torch.load(model_path, map_location=device)
        model.load_state_dict(checkpoint['model_state_dict'])
        model.to(device)
        model.eval()
        
        # Store model info for UI display
        model_info = {
            'path': model_path,
            'type': 'ConvNeXt Large' if 'convnext' in model_path else 'EfficientNet-V2-S',
            'accuracy': checkpoint.get('accuracy', 'Unknown')
        }
        
        health_data = load_json_data('health_data.json')
        allergen_data = load_json_data('allergen_data.json')
        return model, health_data, allergen_data, model_info
    except Exception as e:
        st.error(f"A critical error occurred while loading the model: {e}")
        return None, None, None, None

model, health_data, allergen_data, model_info = load_model_resources()

def transform_image(image_bytes):
    transform = transforms.Compose([
        transforms.Resize((224, 224)), transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])
    image = Image.open(io.BytesIO(image_bytes)).convert("RGB")
    return transform(image).unsqueeze(0)

def get_prediction(image_tensor):
    if model is None: return "Error: Model not loaded", 0.0
    with torch.no_grad():
        outputs = model(image_tensor.to(torch.device('cuda' if torch.cuda.is_available() else 'cpu')))
        probabilities = torch.nn.functional.softmax(outputs, dim=1)
        confidence, predicted_idx = torch.max(probabilities, 1)
        predicted_idx_item = predicted_idx.item()
        if predicted_idx_item >= len(CLASS_NAMES):
            return "Prediction Error: Index out of bounds.", 0.0
        predicted_class = CLASS_NAMES[predicted_idx_item].replace('_', ' ').title()
        return predicted_class, confidence.item()

# --- UI Layout ---
st.markdown("""
<div style="text-align: center; padding: 20px 0;">
    <h1 style="font-size: 3em; margin: 0;">
        🍽️ <span style="color: #28a745;">Eat</span><span style="color: #dc3545;">Smart</span> 
        <span style="color: #17a2b8;">Pro</span>
    </h1>
    <p style="font-size: 1.2em; color: #6c757d; margin: 10px 0;">
        🌟 Your <span style="color: #28a745;">AI-Powered</span> 
        <span style="color: #dc3545;">Nutrition</span> Assistant 🌟
    </p>
    <div style="display: flex; justify-content: center; gap: 10px; margin: 15px 0;">
        <span style="background: linear-gradient(45deg, #28a745, #20c997); color: white; padding: 5px 15px; border-radius: 20px; font-size: 0.9em;">
            🥗 Healthy Analysis
        </span>
        <span style="background: linear-gradient(45deg, #dc3545, #fd7e14); color: white; padding: 5px 15px; border-radius: 20px; font-size: 0.9em;">
            ⚠️ Allergen Alerts
        </span>
        <span style="background: linear-gradient(45deg, #17a2b8, #6f42c1); color: white; padding: 5px 15px; border-radius: 20px; font-size: 0.9em;">
            🍳 Smart Recipes
        </span>
    </div>
</div>
""", unsafe_allow_html=True)

# Model Status Indicator
if model_info:
    model_type = model_info['type']
    model_accuracy = model_info['accuracy']
    
    if 'ConvNeXt' in model_type:
        status_color = "#28a745"  # Green for new model
        status_bg = "#d4edda"
        status_icon = "🚀"
        status_text = f"HIGH ACCURACY MODEL ACTIVE"
    else:
        status_color = "#ffc107"  # Yellow for old model
        status_bg = "#fff3cd"
        status_icon = "⚠️"
        status_text = f"FALLBACK MODEL (Training in Progress)"
    
    st.markdown(f"""
    <div style="background-color: {status_bg}; border: 2px solid {status_color}; border-radius: 10px; padding: 15px; margin: 15px 0; text-align: center;">
        <div style="color: {status_color}; font-size: 1.2em; font-weight: bold;">
            {status_icon} {status_text}
        </div>
        <div style="color: #495057; font-size: 0.9em; margin-top: 5px;">
            Model: {model_type} | Validation Accuracy: {model_accuracy}
        </div>
    </div>
    """, unsafe_allow_html=True)

# User Preferences Sidebar
with st.sidebar:
    st.markdown("""
    <div style="text-align: center; padding: 15px; background: linear-gradient(135deg, #667eea 0%, #764ba2 100%); border-radius: 10px; margin-bottom: 20px;">
        <h3 style="color: white; margin: 0;">⚙️ Your Preferences</h3>
        <p style="color: #f8f9fa; margin: 5px 0; font-size: 0.9em;">Set your dietary needs</p>
    </div>
    """, unsafe_allow_html=True)
    
    # Initialize session state for preferences
    if 'user_allergens' not in st.session_state:
        st.session_state.user_allergens = []
    if 'avoid_trans_fat' not in st.session_state:
        st.session_state.avoid_trans_fat = False
    if 'dietary_preferences' not in st.session_state:
        st.session_state.dietary_preferences = []
    
    # Allergen Preferences
    st.markdown("### 🚨 Allergen Alerts")
    st.markdown("*Select allergens you want to be warned about:*")
    
    common_allergens = ["Gluten", "Dairy", "Egg", "Fish", "Shellfish", "Nuts", "Peanuts", "Soy", "Sesame"]
    
    for allergen in common_allergens:
        if st.checkbox(f"🛡️ {allergen}", key=f"allergen_{allergen}", 
                      value=allergen in st.session_state.user_allergens):
            if allergen not in st.session_state.user_allergens:
                st.session_state.user_allergens.append(allergen)
        else:
            if allergen in st.session_state.user_allergens:
                st.session_state.user_allergens.remove(allergen)
    
    # Trans Fat Preference
    st.markdown("### 🧪 Trans Fat Settings")
    st.session_state.avoid_trans_fat = st.checkbox(
        "⚠️ Alert me about trans fats", 
        value=st.session_state.avoid_trans_fat,
        help="Get warnings about foods that may contain trans fats"
    )
    
    # Dietary Preferences
    st.markdown("### 🌱 Dietary Preferences")
    dietary_options = ["Vegetarian", "Vegan", "Keto", "Low-Carb", "High-Protein", "Gluten-Free"]
    
    for diet in dietary_options:
        if st.checkbox(f"🥬 {diet}", key=f"diet_{diet}", 
                      value=diet in st.session_state.dietary_preferences):
            if diet not in st.session_state.dietary_preferences:
                st.session_state.dietary_preferences.append(diet)
        else:
            if diet in st.session_state.dietary_preferences:
                st.session_state.dietary_preferences.remove(diet)
    
    # Display current preferences summary
    if st.session_state.user_allergens or st.session_state.dietary_preferences or st.session_state.avoid_trans_fat:
        st.markdown("---")
        st.markdown("### 📋 Active Preferences")
        if st.session_state.user_allergens:
            st.markdown(f"🚨 **Allergen Alerts:** {', '.join(st.session_state.user_allergens)}")
        if st.session_state.dietary_preferences:
            st.markdown(f"🌱 **Diet:** {', '.join(st.session_state.dietary_preferences)}")
        if st.session_state.avoid_trans_fat:
            st.markdown("🧪 **Trans Fat Alerts:** Enabled")

if 'image_buffer' not in st.session_state: st.session_state.image_buffer = None
if 'prediction_result' not in st.session_state: st.session_state.prediction_result = None
if 'last_image_buffer' not in st.session_state: st.session_state.last_image_buffer = None

col1, col2 = st.columns([1, 1.2])
with col1:
    st.markdown("""
    <div style="text-align: center; padding: 15px; background: linear-gradient(135deg, #74b9ff 0%, #0984e3 100%); border-radius: 10px; margin-bottom: 20px;">
        <h3 style="color: white; margin: 0;">📸 Upload Food Image</h3>
        <p style="color: #ddd; margin: 5px 0; font-size: 0.9em;">Drag & drop or browse to analyze</p>
    </div>
    """, unsafe_allow_html=True)
    
    uploaded_file = st.file_uploader("Choose your food image", type=["jpg", "jpeg", "png"], label_visibility="collapsed")
    
    # Set the image buffer based on the file uploader's state.
    if uploaded_file is not None:
        st.session_state.image_buffer = uploaded_file.getvalue()
    else:
        st.session_state.image_buffer = None

    # This code block displays the image after it is uploaded.
    if st.session_state.image_buffer is not None:
        st.image(st.session_state.image_buffer, caption='🍽️ Your Food Image', use_column_width=True)

with col2:
    st.markdown("""
    <div style="text-align: center; padding: 15px; background: linear-gradient(135deg, #00b894 0%, #00a085 100%); border-radius: 10px; margin-bottom: 20px;">
        <h3 style="color: white; margin: 0;">🔬 Smart Analysis & Recipes</h3>
        <p style="color: #ddd; margin: 5px 0; font-size: 0.9em;">AI-powered nutrition insights</p>
    </div>
    """, unsafe_allow_html=True)
    
    if model and st.session_state.image_buffer:
        if st.session_state.image_buffer != st.session_state.last_image_buffer:
            st.session_state.last_image_buffer = st.session_state.image_buffer
            with st.spinner('Analyzing image...'):
                image_tensor = transform_image(st.session_state.image_buffer)
                st.session_state.prediction_result = get_prediction(image_tensor)
                if 'recipe' in st.session_state: del st.session_state.recipe
        if st.session_state.prediction_result:
            food_name, confidence = st.session_state.prediction_result
            st.metric(label="Predicted Food", value=food_name)
            st.progress(confidence, text=f"Confidence: {confidence:.2%}")
            tab1, tab2, tab3, tab4 = st.tabs(["Health Info", "Allergen Alert", "Trans Fat Analysis", "AI Recipes"])
            with tab1: 
                health_info_html = get_health_info(food_name, health_data)
                if "BENEFITS_SECTION" in health_info_html:
                    # Display HTML parts
                    st.markdown(health_info_html.replace("BENEFITS_SECTION", ""), unsafe_allow_html=True)
                    
                    # Display benefits using Streamlit components for proper formatting
                    nutrition_info = health_data.get("nutrition_info", {})
                    food_info = nutrition_info.get(food_name.replace('_', ' ').title()) or nutrition_info.get(food_name.lower()) or health_data.get(food_name.replace('_', ' ').title())
                    
                    if food_info and food_info.get("benefits"):
                        st.markdown("### ✨ Health Benefits")
                        for benefit in food_info.get("benefits", []):
                            st.markdown(f"• {benefit}")
                else:
                    st.markdown(health_info_html, unsafe_allow_html=True)
            with tab2: 
                get_allergen_info(food_name, allergen_data)
            with tab3:
                st.markdown(get_trans_fat_analysis(food_name, health_data), unsafe_allow_html=True)
            with tab4:
                st.subheader(f"AI-Generated Recipe for {food_name}")
                if 'recipe' not in st.session_state or st.session_state.get('recipe_food') != food_name:
                    with st.spinner("Chef AI is thinking of a recipe..."):
                        st.session_state.recipe = generate_recipe(food_name)
                        st.session_state.recipe_food = food_name
                if 'recipe' in st.session_state:
                    recipe_text = st.session_state.recipe
                    sections = {"ingredients": [], "instructions": [], "tips": []}
                    current_section_key = None
                    for line in recipe_text.split('\n'):
                        line_lower = line.strip().lower()
                        if line_lower.startswith("ingredients"): current_section_key = "ingredients"
                        elif line_lower.startswith("instructions"): current_section_key = "instructions"
                        elif line_lower.startswith("tips") or line_lower.startswith("chef's tips"): current_section_key = "tips"
                        elif line.strip() and current_section_key: sections[current_section_key].append(line.strip().lstrip('*- '))
                    with st.expander("Ingredients", expanded=True): st.markdown("\n".join(f"- {item}" for item in sections["ingredients"]) or "No ingredients listed.")
                    with st.expander("Instructions", expanded=True): st.markdown("\n".join(f"{i+1}. {item}" for i, item in enumerate(sections["instructions"])) or "No instructions provided.")
                    with st.expander("Chef's Tips"): st.markdown("\n".join(f"- {item}" for item in sections["tips"]) or "No special tips provided.")
    elif not model:
        st.error("Application has failed to start. Please check the logs for errors.")
    else: 
        st.info("Upload an image to get started.")