Do I Need to Split My Dataset for Training GANs?

Hello everyone,

I’m relatively new to training Generative Adversarial Networks (GANs) and have a question regarding dataset preparation. In traditional supervised learning, I’m used to splitting my data into training and test sets (and sometimes a validation set). However, I’m uncertain if this same approach is necessary or recommended for GANs.

Do I need to divide my dataset when training GANs? If so, why, and how should I use the separated data during the training and evaluation process? If not, can you provide insights into why this might not be necessary for GANs?

Any guidance or references on this topic would be greatly appreciated!

Thank you in advance!

At the end of the day, I think what matters is how you train the GAN.
What I mean by the big elephant in the room is the training phase not exactly if you split in 20% for the test and the rest for training (you can do that, to evaluate, but still the big elephant is how you achieve a good training). Again, this is just my opinion.