pasztorb / rotational_cad

Rotational_CAD

This github repository was made for my project that concerns autoencoders.

CAD.py

This is the main script that includes the models and their training. It can be run from terminal with the following command.

python3 CAD.py model_type input_data_path output_path_to_directory

The first argument relates to the different types of models I considered. Possible types are: flat, convt, upsample, comb.

preprocess_data.py

This script generates the training data from the MNIST dataset. Running it will generate the data into the working directory.

visualize.py

This script plots the predicted images side-by-side with the input and the target images. Sample running command:

python3 visualize.py path_to_model output_path_to_directory

Upon calling it will generate 20 images from the test, validation and training images each i.e. 60 overall.

evaluation.ipynb

This is an iPython notebook I used to evaluate the test performance of the models and to examine the samples with the worst MSE results.

test_evaluate.py

This script simply calculates the test error for the given model.