RepAug 0.0.1

A package for reproducible vision data augmentation in PyTorch.

RepAug

Reproducible vision data augmentation in PyTorch.

Essentially, images are converted to torch.Tensor first to be transformed. You can specify the argument seed, which will be passed to np.random.default_rng, to make your transform reproducible.

Currently supported transforms:

• RandomColorJitter
• RandomCrop, RandomResizedCrop
• RandomHorizontalFlip, RandomVerticalFlip
• RandomGaussianBlur
• Salt, Pepper
• RandomRotation, Random90Rotation, Random180Rotation

See illustration.ipynb for illustrations.

To use this package, you can clone it first and run python setup.py install.