I am wondering how much I should extend my training set with data augmentation. Is there somewhere a pre-defined number I can go with?

Suppose I have 10000 images, can I go as far as 10x or 20x times, to get 100000 and 200000, respectively, images? I am wondering how will this impact model training. I am using a mask R-CNN.

  • 1
    $\begingroup$ If hardware and training time isn't a limitation, just try it and see when validation accuracy starts to decrease. $\endgroup$ – JobHunter69 Sep 8 '20 at 14:21
  • $\begingroup$ Maybe you should describe which model you are trying to use (I suppose it's a neural network) and how you are augmenting your training data. There are several ways you can augment the training data. $\endgroup$ – nbro Sep 9 '20 at 10:34
  • $\begingroup$ @nbro Mask RCNN, this is the model. $\endgroup$ – Ad Blu Sep 9 '20 at 12:29
  • $\begingroup$ Literally augment as much as possible as long as a valid image is still produced. The gold standard for CV is human equivalent understanding. Now you dont need to preagument them, they can be augmented at sampling time. $\endgroup$ – FourierFlux Sep 9 '20 at 23:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.