CODEWITHSUNDEEP
pythontensorflow2.0data-augmentation
newbee1905
newbee1905

Reputation: 1

Should Augmented Data be Added to Existing Data or Used as Complete Replacement in CNN Models?

I need help with the optimal approach for integrating augmented data into Convolutional Neural Network (CNN) models. Specifically, should augmented data be added to the existing dataset to expand its size, or should it replace the original data entirely? I would appreciate any advice, best practices, or experiences regarding this matter. Thank you!

I checked the tutorials about Data Augmentation on the TensorFlow website, they just make it Sequential and either use it as a part of the layer in the model or replace the original dataset with the augmented one. However, one of the reasons for augmenting data I found is for smaller datasets. So wouldn't adding augmented data to the original one would be more suitable for that benefit? Or is augmenting data to make the model able to get more meaningful features which results in better results even with less data?

Upvotes: -1

Views: 178

Answers (1)

newbee1905
newbee1905

Reputation: 1

I have found the answer to my question. After rechecking how ImageDataGenerator works in TensorFlow, it will create a new augmented version of the images for each epoch. If I understand it correctly, it would be adding more pictures to the existing data while training, instead of directly doing it.

Upvotes: 0

Related Questions