import gradio as gr
import torch
from torch import nn
from torchvision import models, transforms
from PIL import Image
from transformers import AutoFeatureExtractor

# Load the model from Hugging Face model hub
model_id = "KabeerAmjad/food_classification_model"
# Load ResNet50 model and adjust the final layer
model = models.resnet50(pretrained=True)
model.fc = nn.Linear(model.fc.in_features, 11)  # Adjust the output layer to match your number of classes

# Load the weights from the Hugging Face model hub
model.load_state_dict(torch.hub.load_state_dict_from_url(f"https://huggingface.co/{model_id}/resolve/main/food_classification_model.pth"))
model.eval()

# Load the feature extractor
feature_extractor = AutoFeatureExtractor.from_pretrained(model_id)

# Define the same preprocessing used during training
transform = transforms.Compose([
    transforms.Resize((224, 224)),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

# Define the prediction function
def classify_image(img):
    # Preprocess the image
    img = transform(img).unsqueeze(0)  # Add batch dimension
    
    # Make prediction
    with torch.no_grad():
        outputs = model(img)
        probs = torch.softmax(outputs, dim=-1)
    
    # Get the label with the highest probability
    top_label = model.config.id2label[probs.argmax().item()]  # Map to label (use your custom label mapping if needed)
    return top_label

# Create the Gradio interface
iface = gr.Interface(
    fn=classify_image,
    inputs=gr.Image(type="pil"),
    outputs="text",
    title="Food Image Classification",
    description="Upload an image to classify if it’s an apple pie, etc."
)

# Launch the app
iface.launch()