When we use the term of Data Augmentation, I believe that the images like x and rotated x are two different images that will be fed to model. Indeed, when we have a dataset of n images, we will have d*n images with Data Augmentation which d is the number of transformations and their combinations . However, when we use Transform.Compose, it transform any image to one image.
How we can apply range of transformation like resize, rotate, crop ā¦ to list of images to get d*n which d in the number of combination of transformation?
Iām not sure if you are looking for a way to create all possible transformations defined in Compose (this would be quite impossible) or e.g. 5 outputs for 5 chained transformations.
In both cases you can just call out = transform(x) multiple times since the transformations will be applied on the fly.
1 Like
I thought that we should do this however, I understand that the number of data will increase exponentially !
Thanks