import gradio as gr
import torch
from transformers import ViTFeatureExtractor, ViTForImageClassification
from PIL import Image
import numpy as np
import rembg

# Define the model and feature extractor
model_name ="KhadijaAsehnoune12/ViTOrangeLeafDiseaseClassifier"
model = ViTForImageClassification.from_pretrained(model_name, num_labels=10, ignore_mismatched_sizes=True)
feature_extractor = ViTFeatureExtractor.from_pretrained(model_name)

# Define the label mapping
id2label = {
    "0": "Aleurocanthus spiniferus",
    "1": "Chancre citrique",
    "2": "Cochenille blanche",
    "3": "Dépérissement des agrumes",
    "4": "Feuille saine",
    "5": "Jaunissement des feuilles",
    "6": "Maladie de l'oïdium",
    "7": "Maladie du dragon jaune",
    "8": "Mineuse des agrumes",
    "9": "Trou de balle"
}

def remove_background(image):
    # Convert the image to RGBA
    image = image.convert("RGBA")
    
    # Remove the background
    image_np = np.array(image)
    output_np = rembg.remove(image_np)
    
    # Create a white background image
    white_bg = Image.new("RGBA", image.size, "WHITE")
    
    # Composite the original image over the white background
    output_image = Image.alpha_composite(white_bg, Image.fromarray(output_np))
    
    # Convert back to RGB
    output_image = output_image.convert("RGB")
    
    return output_image

def predict(image):
    # Remove the background
    image = remove_background(image)
    
    # Preprocess the image
    inputs = feature_extractor(images=image, return_tensors="pt")
    
    # Forward pass through the model
    outputs = model(**inputs)
    
    # Get the logits
    logits = outputs.logits
    
    # Calculate confidence scores with softmax
    probs = torch.nn.functional.softmax(logits, dim=-1)[0]
    
    # Get the index of the most probable class
    predicted_class_idx = probs.argmax().item()
    
    # Get the label and confidence score of the most probable class
    predicted_label = id2label[str(predicted_class_idx)]
    confidence_score = probs[predicted_class_idx].item() * 100  # Multiply by 100 to get a percentage
    
    # Return the label and confidence score
    return f"{predicted_label}: {confidence_score:.2f}%"

# Create the Gradio interface
image = gr.Image(type="pil") 
label = gr.Textbox(label="Prediction")

gr.Interface(fn=predict,
             inputs=image,
             outputs=label,
             title="Classification des maladies des agrumes",
             description="Téléchargez une image d'une feuille d'agrume pour classer sa maladie. Le modèle est entraîné sur les maladies suivantes : Aleurocanthus spiniferus, Chancre citrique, Cochenille blanche, Dépérissement des agrumes, Feuille saine, Jaunissement des feuilles, Maladie de l'oïdium, Maladie du dragon jaune, Mineuse des agrumes, Trou de balle.",
             examples=["maladie_du_dragon_jaune.jpg", "critique.jpg", "feuille_saine.jpg"]).launch(share=True)