##########################################################

To run the experiments in section 4.3

First download the curated metfaces data and place it in this folder: https://github.com/NVlabs/metfaces-dataset

Next downsample these images using:

python downsample.py

Then create random labels and split into train and test using:

python split_train_test.py

This should create a test.csv and train.csv

Next train the model using:

python train.py <seed>

Do this for multiple seeds. All model pkl files will be stored in metfaces_models/

Now download the image curves here: https://drive.google.com/file/d/1x5t-sc9NlW5N_ZBXUM0WcfX-toUXa85L/view?usp=sharing

And place it in the root directory of the project.

Finally, get the number of linear regions on the curve (blue) using:

python images_find_linear_regions.py <seed> <path/to/root/cifar-out/out1>

Do the above for all the out directories (1 to 8)

Finally, get the number of linear regions on the line (green) using:

python images_find_linear_regions.py <seed> <path/to/root/cifar-out/out1>

Do the above for all the out directories (1 to 8)

Finally, replace the root folder variable in analyse_data.py to run the code that generates all the graphs.

##########################################################
