Differentiable Data Augmentation.

facebookresearch / radioactive_data

This technique modifies image data so that any model trained on it will bear an identifiable mark.

Other

39 stars 9 forks source link

Differentiable Data Augmentation. #1

Closed MaFuyan closed 4 years ago

MaFuyan commented 4 years ago

Hi, buddy. I really appreciate it that you make your brilliant work open-sourced.

After diving into your paper, I thought the Differentiable Data Augmentation module was pretty interesting. However, I found it was not implemented in src/data_augmentations/init.py. Would you mind sharing how you implement that part?

Thanks in advance.

alexandresablayrolles commented 4 years ago

All the code should be in src/data_augmentations/init.py, with the class DifferentiableDataAugmentation and its children CenterCrop and RandomResizedCropFlip.