Hi, I am unsure of one thing. I made k-fold cross validation on my entire dataset of brain tumors MRI image classification task and I used argumentation so that only my current training part in each fold is augmented.
Now the question is, when I finalize the model idea, because there seems to not be a lot overfitting in k-fold cross validation, as far as I understand I can use entire dataset (that means take all the images from training and test directories and use every image I have) for retraining of the final model and then based on its predictions I would report accuracy and loss?
Also, during retraining of the final model, should the images be augmented in the same way as they were in k-fold cross-validation, or I shouldn’t augment them?
Hi @daniel11, since you have applied augmentation while performing k-fold cross validation and have observed less overfitting you can apply the data augmentation in the final training as well. Thank You.