app.py

import streamlit as st
import torch
from model import Generator
import torchvision.utils as vutils
import os
from math import log2

# Function to generate images
def generate_images():
    Z_DIM = 256
    IN_CHANNELS = 256

    # Load pretrained generator weights
    checkpoint = torch.load("generator.pth", map_location=torch.device('cpu'))

    # Filter out optimizer-related keys
    state_dict = checkpoint['state_dict']

    # Load the filtered state dictionary into the model
    generator = Generator(Z_DIM, IN_CHANNELS, img_channels=3)
    generator.load_state_dict(state_dict)
    generator.eval()

    # Set output directory
    output_dir = "generated_images"
    os.makedirs(output_dir, exist_ok=True)

    # Generate images
    img_sizes = [256]
    images = []
    for img_size in img_sizes:
        num_steps = int(log2(img_size / 4))
        x = torch.randn((6, Z_DIM, 1, 1))  # Generate a batch of 6 images
        with torch.no_grad():
            z = generator(x, alpha=0.5, steps=num_steps)

        # Normalize the generated images to the range [-1, 1]
        z = (z + 1) / 2

        assert z.shape == (6, 3, img_size, img_size)

        # Append generated images to the list
        for i in range(6):
            images.append(z[i].detach())

    return images

# Main function to create Streamlit web app
def main():
    st.title('Image Generation with pro-gan 🤖')
    st.write("Click the buttons below to generate images.")
    st.write("Trained on CelebHQ dataset.")

    # Prompt message about image size
    st.write("Note: Due to limited resources, the model has been trained to generate 256x256 size images. They are still awesome!")

    # Generate images on button click
    if st.button('Generate Images'):
        images = generate_images()
        # Display the generated images
        for i, image in enumerate(images):
            st.image(image.permute(1, 2, 0).cpu().numpy(), caption=f'Generated Image {i+1}', use_column_width=True)

if __name__ == '__main__':
    main()