import os
import torch
import torchvision.transforms as transforms
import torchvision.models as models
from PIL import Image
import json
import gradio as gr
import requests

# Path to the model file and Hugging Face URL
model_path = 'food_classification_model.pth'
model_url = "https://huggingface.co/KabeerAmjad/food_classification_model/resolve/main/food_classification_model.pth"

# Download the model file if it's not already available
if not os.path.exists(model_path):
    print(f"Downloading the model from {model_url}...")
    response = requests.get(model_url)
    with open(model_path, 'wb') as f:
        f.write(response.content)
    print("Model downloaded successfully.")

# Load the model with updated weights parameter
model = models.resnet50(weights=models.ResNet50_Weights.DEFAULT)
model.eval()  # Set model to evaluation mode

# Load the model's custom state_dict
try:
    state_dict = torch.load(model_path, map_location=torch.device('cpu'))
    model.load_state_dict(state_dict)
except RuntimeError as e:
    print("Error loading state_dict:", e)
    print("Ensure that the saved model architecture matches ResNet50.")

# Define the image transformations
preprocess = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225],
    ),
])

# Load labels
with open("config.json") as f:
    labels = json.load(f)

# Function to predict image class
def predict(image):
    # Convert the uploaded file to a PIL image
    input_image = image.convert("RGB")
    
    # Preprocess the image
    input_tensor = preprocess(input_image)
    input_batch = input_tensor.unsqueeze(0)  # Add batch dimension
    
    # Check if a GPU is available and move the input and model to GPU
    if torch.cuda.is_available():
        input_batch = input_batch.to('cuda')
        model.to('cuda')
    
    # Perform inference
    with torch.no_grad():
        output = model(input_batch)
    
    # Get the predicted class with the highest score
    _, predicted_idx = torch.max(output, 1)
    predicted_class = labels[str(predicted_idx.item())]

    return f"Predicted class: {predicted_class}"

# Set up the Gradio interface
iface = gr.Interface(
    fn=predict,
    inputs=gr.Image(type="pil"),
    outputs="text",
    title="Food Classification Model",
    description="Upload an image of food to classify it."
)

# Launch the Gradio app
iface.launch()