Compare multiple TF Hub model performance relative to images
This file compares a few models to images from wiki to check whether it "holds up" to the claim that every improvement is significantly better than the previous.
This repo served to be a random "eureka" moment for myself that bear fruit (albeit roughy).
It took me just a few days to slap together some code with the modified official tfhub example.
A mentor advised that I can write a paper with this, but I'm full of tasks so instead I'll just throw this up here.
I might try to in the future(TM), but in the meantime feel free to use that script. It'd be nice if I get a mention but it is not required.