Self-Supervised Learning with Data Augmentations Provably Isolates Content from Style [NeurIPS 2021] Official code to reproduce the results and data p