michalsr / ohl_auto_aug

Online Hyper-parameter Learning for Auto-Augmentation

Implementation of CIFAR-10 experiments in [Online Hyper-parameter Learning for Auto-Augmentation Strategy] (https://arxiv.org/abs/1905.07373) from ICCV 2019

Dependencies

• PyTorch 1.8
• TorchVision

Code

To run the CIFAR-10 experiments: python main.py --output_dir {OUTPUT directory} --prefix {directory before main directoroy name} --auto_aug {boolean for whether to use normal augmentations or auto-augmentations} --epoch {number of total epochs} --trajec {number of trajectories}