how does 'DataAugmentation' option in pixelLabelImageDatastore work?

Hello,

I don't understand how 'pixelLabelDatastore' works when training a U-net using augmented data. I know that the code is the following:

ds = pixelLabelImageSource(imds,pxds,'DataAugmentation',imageAugmenter);

net = trainNetwork(ds,lgraph,options);

Is the U-net trained now using augmented data, or are the augmented images used only when calculating the loss in the mini-batches (and not for training)?

Thanks in advance for any answer!

Answer 1

[Shivam Sardana]

As mentioned in pixelLabelDatastore documentation, ‘DatatAugemtation’ parameter augments data during training. So, U-net is trained on augmented data now.