To reproduce the figures & tables from the paper, run the paper_figures.ipynb notebook. You can also visualize the notebook using NBViewer here.
The raw results used as a basis for the paper are in the result_csvs folder. These CSV files were generated from update_allresults_file.py, which loads the .pkl files of individual models/OOD detectors from ../results/ and aggregates them into a Pandas dataframe.