utils.py

import idx2numpy, torch
import torch
import torch.nn as nn
import torch.nn.functional as F
from torchvision import transforms, datasets
from PIL import Image


class ApplyEnhancementFilter:
	def __init__(self, out_channels, kernel_size, stride=1, padding=0, bias=False):
		"""
		Initialize the convolution parameters.
		"""
		self.out_channels = out_channels
		self.kernel_size = kernel_size
		self.stride = stride
		self.padding = padding
		self.bias = bias
		# Define the convolutional layer (not trained here)
		self.conv = nn.Conv2d(in_channels=1,  # Adjust this based on your image channels (1 for grayscale, 3 for RGB)
		                      out_channels=out_channels,
		                      kernel_size=kernel_size,
		                      stride=stride,
		                      padding=padding,
		                      bias=bias)

		# Example: Manually defining a simple edge-detection kernel
		# For a real use-case, the kernel weights would be learned or defined according to the filter you need.
		edge_detection_kernel = torch.tensor([[0, -1., 0.],
		                                      [-1., 5., -1.],
		                                      [0., -1., 0.]]).unsqueeze(0).unsqueeze(0)
		self.conv.weight = nn.Parameter(edge_detection_kernel.float())

	def __call__(self, img):
		"""
		Apply the convolution transformation.
		"""
		# Convert PIL image to tensor
		img_tensor = transforms.functional.to_tensor(img).unsqueeze(0)  # Add batch dimension
		# Apply convolution
		conv_img = self.conv(img_tensor)
		# Remove batch dimension and convert back to PIL image for further transformations or visualization
		conv_img_pil = transforms.functional.to_pil_image(conv_img.squeeze(0))
		return conv_img_pil


if __name__ == "__main__":
	# It is important to normalise the dataset, so no specific input effects the model more than other based purely on input values.
	# As values can range from 0-255, this can cause problems, so z-score will be used via Transforms.

	# First we need the mean and standard deviation of train dataset.

	train_images = idx2numpy.convert_from_file("mnist_dataset/train-images.idx3-ubyte")

	# Convert the training images to a PyTorch tensor and scale values to [0, 1]
	train_images_tensor = torch.tensor(train_images, dtype=torch.float32) / 255.0

	train_mean = train_images_tensor.mean()
	train_std = train_images_tensor.std()

	print(f"Mean: {train_mean}, Std: {train_std}")
	# Mean: 0.13066047430038452, Std: 0.30810782313346863