Based on the official implementation of Fast AutoAugment but runs on a single GPU, uses the newest version of Ray, and is much easier to get running.
- Fast AutoAugment learns augmentation policies using a more efficient search strategy based on density matching.
- Fast AutoAugment speeds up the search time by orders of magnitude while maintaining the comparable performances.