ImageDenoisingAutoencoder

The Denoising autoencoder is implemented with Pytorch and is applied on the MNIST dataset.

Table of contents

• Required Imports
• Dataset
• Denoising Autoencoder
• Model architecture
• Hyperparameters
• Training
• Training-Loss plot
• Results

Required Imports

• NumPy
• Matplotlib(for data visualization)
• PyTorch(for building and training the autoencoder)

Dataset

The MNIST Dataset is used. It is a dataset of 60,000 training samples and 10,000 test samples. Each sample is a 28×28 pixel grayscale image of a single handwritten digit between 0 and 9.

Denoising Autoencoder

• Components of Autoencoder:
  • Encoder - This network downsamples the data into lower dimensions.
  • Decoder - This network reconstructs the original data from the lower dimensional representation
• In denoising autoencoders some noise is introduced to the input images. It tries to reconstruct the original image without noise. Such a noisy input reduces the risk of overfitting and prevents the autoencoder from learning a simple identity function.

Model architecture

#Encoder
      nn.Conv2d(1,16,3, stride = 2, padding = 1)
      nn.ReLU()
      nn.Conv2d(16,32,3, stride = 2, padding = 1)
      nn.ReLU()
      nn.Conv2d(32,64,7)
#Decoder
      nn.ConvTranspose2d(64,32,7)
      nn.ReLU()
      nn.ConvTranspose2d(32,16,3, stride = 2, padding=1, output_padding = 1)
      nn.ReLU()
      nn.ConvTranspose2d(16,1,3, stride = 2, padding=1, output_padding = 1)
      nn.Sigmoid()    

Hyperparameters

Hyperparameter	Value
Learning rate	0.001
Number of epochs	20
Batch Size	64

Training

Adding noise

A function is defined to add noise to the images. torch.randn is used to create a noisy tensor of the same size as the input.

Optimizer and Loss function

MSE loss and the Adam optimization technique is used.

Training-Loss plot

Results

  Original                  Noisy                Denoised