If I have training set of 10 images, when I perform get_transforms will I get 10 transformed images, or will it create additional images using transformations, in that case I would get 10 resized + new transformed images? If so, can I somehow tune how much images I want to create?
1 Like
You will get 10 images every time, it will just give you a transformed version of the image every time an image is loaded. So every epoch, the model would in this case read 10 images, then the next epoch, it would read the same 10 images, but they would be re-transformed from the base image again.
3 Likes
I understand it every time there is image request transformations are applied wit defined probability.
1 Like
Thank you for the answers.
When you say probability, probability of what? The reason why I ask this is because I want to create more diverse pictures in each epoch, in a way to increase training set and reduce overfitting.
Do you, by any chance, know can I somehow increase level of data augmentation trough get transforms? I’m working with Pneumonia X-ray pictures.
Hi,
Look in to https://docs.fast.ai/vision.transform.html#Randomness
Well indeed radomnes is your choise if you declare value p
in deterministic function, this function will be used with p-probability/
or you can declare parameter of function in range which will be applied radomnly.
If you don’t declaree p prpbability this transformation is always going to be applied.
having few function in set of transformations with p give you endless combinations.
Hope I made it clearer 
Cheers
Michal