Code release for Unsupervised Domain Adaptation through Self-Supervision.
Our code requires pytorch version 1.0 or higher. install with pip install -r requirements.txt --user

1. Modify the scripts in scripts/. --outf sets the output folder, and --data_root sets the input data folder.
2. Run the scripts in scripts/. Each script corresponds to a column in our results table.
3. show_table.py prints the results and plots the error curves. For ease of comparison, two results are printed: the smallest error and the early stopping error according to our selection heuristics.
4. figure_mmd.py reproduces the two convergence figures shown in our paper.

The datasets used for each command are downloaded automatically to --data_root.