app.py

import gradio as gr
import tensorflow as tf
import numpy as np
import re
import base64
import io
from PIL import Image

# Load the saved model
model = tf.keras.models.load_model('banana_disease_densenet121.keras')

# Load class names
class_names = np.load("class_names.npy", allow_pickle=True)

# Preprocess the input image
def preprocess_image(img):
    if img is None:
        return None
    img = img.resize((256, 256))  # Resize to match training
    img = np.array(img) / 255.0   # Normalize
    img = np.expand_dims(img, axis=0)  # Add batch dimension
    return img

# Prediction function
# Prediction function


def predict_disease(img):
    """
    Predict banana disease from image.
    Handles both PIL images and base64 strings from Flutter.
    Returns filtered results (only Panama & Healthy)
    """
    try:
        # Handle base64 encoded images from Flutter
        if isinstance(img, str):
            # Remove data URL prefix if present (e.g., "data:image/jpeg;base64,")
            if img.startswith('data:image'):
                img = re.sub(r'^data:image/.+;base64,', '', img)
            
            # Decode base64 to image
            image_data = base64.b64decode(img)
            img = Image.open(io.BytesIO(image_data))
            
            # Ensure RGB mode
            if img.mode != 'RGB':
                img = img.convert('RGB')
        
        # Preprocess the image
        img_processed = preprocess_image(img)
        if img_processed is None:
            return "⚠️ No image provided", {}
        
        # Predict
        predictions = model.predict(img_processed)[0]
        predicted_class = np.argmax(predictions)
        
        # Filter to show only Panama Disease and Healthy Leaf
        filtered_classes = ['Banana Panama Disease', 'Banana Healthy Leaf']
        confidence_scores = {
            class_name: float(predictions[i]) 
            for i, class_name in enumerate(class_names) 
            if class_name in filtered_classes
        }
        
        return f"Predicted: {class_names[predicted_class]}", confidence_scores
        
    except Exception as e:
        print(f"Error in prediction: {str(e)}")
        return f"Error: {str(e)}", {}
demo = gr.Interface(
    fn=predict_disease,
    inputs=gr.Image(type="pil", label="Upload Banana Leaf"),
    outputs=[
        gr.Text(label="Prediction"),
        gr.Label(label="Confidence Scores", num_top_classes=2)
    ],
    title="🍌 Banana Leaf Disease Classifier",
    description="Upload a banana leaf image, and our AI will diagnose the disease",
    theme=gr.themes.Soft(),
    api_name="predict"
)



# Launch app direct
demo.launch()