app.py

import torch
from transformers import AutoModel
import torch.nn as nn
from PIL import Image
import numpy as np
import streamlit as st

# Set the device
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

# Load the trained model from the Hugging Face Hub
model = AutoModel.from_pretrained('dhhd255/parkinsons_pred0.1')

# Move the model to the device
model = model.to(device)

# Use Streamlit to upload an image
uploaded_file = st.file_uploader("Choose an image...", type=["png", "jpg", "jpeg"])
if uploaded_file is not None:
    # Load and resize the image
    image_size = (224, 224)
    new_image = Image.open(uploaded_file).convert('RGB').resize(image_size)
    new_image = np.array(new_image)
    new_image = torch.from_numpy(new_image).transpose(0, 2).float().unsqueeze(0)

    # Move the data to the device
    new_image = new_image.to(device)

    # Make predictions using the trained model
    with torch.no_grad():
        predictions = model(new_image)
        logits = predictions.last_hidden_state
        logits = logits.view(logits.shape[0], -1)
        num_classes=2
        feature_reducer = nn.Linear(logits.shape[1], num_classes)

        logits = logits.to(device)
        feature_reducer = feature_reducer.to(device)

        logits = feature_reducer(logits)
        predicted_class = torch.argmax(logits, dim=1).item()
        if(predicted_class == 0):
            st.write('Predicted class: Parkinson\'s')
        else:
            st.write('Predicted class: Healthy')