what does the validation split do? I have found that the number of images remains the same for training and validation even if I do validation split to 0.1 , 0.2 or 0.
And what is the 18 of 88 is representing here in epochs
The validation split is the part of the dataset that you will be using to test your model’s performance. According to how the model performs on the validation and training splits then you change the hyperparameters to improve your models performance.
The 18/88 means that is currently training on batch 18 of 88 batches that your training dataset has been split, because training is performed on batches of examples.
but i already have a validation directory dataset which I will be passing in model.fit , then why do I need to do validation split in datagenerator
That is used for data augmentation in the validattion dataset.
One more thing . Lets say I have 3000 training images in the training set directory . then after applying data augmentation in the image data generator . Will the number of images increase from 3000 . And how do I know how many number of augmented images will be used in the training .
The number in the folder will not increase those images are made on the run and used.
How many thats a good question, a while back in another specialization here I noticed that every image goes through augmentation, and if you have different kind of augmentations then you will have many instances produced. The best way to find out is to check the github repo of tensorflow section of data augmentation or maybe tensorflow itself.