Implementing a Convolutional Autoencoder with PyTorch