Official Tensorflow implementation of U-GAT-IT: Unsupervised Generative Attentional Networks with Adaptive Layer-Instance Normalization for Image-to-Image Translation (ICLR 2020)
With this piece of code, we make sure that the augmentation is provided with a probability of 0.5. In the paper, it is written that the horizontal flip is done with a probability of 0.5 and then the random crops are taken (assuming with a probability of 1, ie all the time).
For data augmentation, we flipped the images horizontally with a probability of 0.5, resized them to 286 x 286, and random cropped them to 256 x 256.
With the implementation of the code, the statement in the paper becomes untrue, here the horizontal flip probability becomes 0.5*0.5=0.25 and the random crop probability becomes 0.5.
https://github.com/taki0112/UGATIT/blob/d508e8f5188e47000d79d8aecada0cc9119e0d56/utils.py#L23-L24
With this piece of code, we make sure that the augmentation is provided with a probability of
0.5
. In the paper, it is written that the horizontal flip is done with a probability of0.5
and then the random crops are taken (assuming with a probability of1
, ie all the time).With the implementation of the code, the statement in the paper becomes untrue, here the horizontal flip probability becomes
0.5*0.5=0.25
and the random crop probability becomes0.5
.Is this intended behaviour?