CIFAR image augmentation


(Naga Supraja) #1

Hi, I am just starting up on deep learning. I want to start working on the CIFAR dataset. I believe we have to enable data augmentation while training. I was just wondering, by enabling data augmentation, how many total images does the network “see” before it can classify well enough.
Thanks.


#2

Hi there,
First off, congrats on starting off!

With regards to your question, I would recommend trying and seeing! You can monitor the performance during training by looking at withheld validation data. This should be able to inform you.

I can’t really answer how many images it needs to see before it works – it will depend on network architecture, learning rate and so on.

You should give it a try and see what you get!