image reconstruction

this repository contains two examples of reconstructing input images/videos with deep learning methods using pyTorch.

Auto Encoder

A simple auto encoder as described here:

it maps the images into the code, and then maps the code to the reconstructed images.

Input images are on the first row.

first example of input and output digits from the MNIST dataset

second example of input and output digits from the MNIST dataset

dataset: MNIST

a simple, stacked Reccurent Neural Network model from pyTorch with ReLU activation function.

the loss function (mean square error) is minimized using Adam optimization.

first example of input and output frames from the moving MNIST dataset

second example of input and output frames from the moving MNIST dataset

This script contains a class for easy loading of data from the moving MNIST dataset.

The dataset is splitted in training-validation and testing subsets.
The training set is being shuffled after loading all data from it.

code for displaying images using matplotlib

For every method, the properties of the model are in the file model“method name”.py and the rest under the train“method name”.py