Running on your own data

[Sicily-F]

Hi there,

I haven't tested this repo yet, but I presume it would be something along the lines of just respecifying the number of classes and images, as well as your directory to images? I've not seen the model up on Tensorflow Hub yet, any ideas would be great!

[abhay-7]

I tried running this on my own data, but somehow it is just predicting a single class as the output. Any tips to debug this issue? (I have already tried playing around with the activation functions in the final and intermediate layers, and I am using one other dense layer over and above the nfnet model defined in nfnet.py)