jramapuram / datasets Goto Github PK

I use the code, but the prob 0.2 is not for meaning transform gray scale in 0.2 prob.

First, appreciate your codes that have helped me a lot. However, I may find a mistake in your codes, please let me know if I am wrong.

Since DALI does not support the condition augmentation, I know that you have to use mixing to achieve such operations in

According to the docs of pytorch, the augmentations are conducted with a prob p. In other words, the p for 1 that augmented the images.

Given a simple example like transforms.RandomHorizontalFlip. Idealy, zero prob will do not ever flip the images and and vice versa.

I have tried your code, but I found it performs contrary to pytorch.

Also, torch randomly augmented with a normal distribution like https://pytorch.org/vision/stable/_modules/torchvision/transforms/transforms.html#RandomGrayscale, instead of uniform.

That's all, looking forward to your reply.