utils.py

import torch
import numpy as np
from PIL import Image
from torchvision.transforms import Compose, Resize, Grayscale, ToTensor, ToPILImage


# global variable for the grayscale transform
transform_gs = Compose(
    [Resize((128, 128)), Grayscale(num_output_channels=1), ToTensor()]
)

def process_gs_image(image):
    """
    Function to process the grayscale image.
    """
    # Save original size for later use
    original_size = image.size  # (width, height)
    
    # Convert the image to grayscale and resize
    image = transform_gs(image)
    
    # Add the batch dimension
    image = image.unsqueeze(0)
    
    # Return both the processed image and original size
    return image, original_size

def inverse_transform_cs(tensor, original_size):
    """
    Function to convert the tensor back to the color image and resize it to its original size.
    """
    # Convert the tensor back to a PIL image
    to_pil = ToPILImage()
    pil_image = to_pil(tensor.squeeze(0))  # Remove the batch dimension
    
    # Resize the image back to the original size
    pil_image = pil_image.resize(original_size)
    
    return pil_image