I am following this tutorial about pix2pix Conditional GAN.
I have a doubt regarding the Data Augmentation step.
In the notebook, author uses transformation function (load_image_train and load_image_test) directly on the dataset and use that dataset for building the model.
My Question is : How is that useful, as per my understanding, this would simply modify each image in the dataset with “random_jittered” image and throughout training nothing in dataset would change?
So when during the training the augmentation (or benifit of augmentation) is achieved.
Does it have something to do with @tf.function() decorator on function random_jitter(input_image, real_image) ?.
random_jitter function is applied only to the load_image_train. This transformation function adds more images to the training dataset. Here there is no significance of @tf.function decorator with random_jitter. Thank you
Thanks for the response.
This transformation function adds more images to the training dataset.
But isn’t load_image_train a one-to-one function, taking a File Path, decoding input and real images, transforming them and returning them?
I checked the cardinality of dataset before and after performing dataset.map and both are 400.