TF (or Keras via TF) is definitely the most popular framework for newer projects, although some argue PyTorch is holistically better.
I think pytorch is more popular in research, but TF is the most popular for commercial applications right now
I have to use TF at work (small startup), but I use Pytorch for personal projects and research.
Re:Tensorflow 2.0.0-Alpha0
by FridgeSeal on Wednesday March 06, 2019 @06:33PM
(#19324091)
I'm about to use Flux.jl for my next project.
I have previously tried using TF but it was super painful so I used Keras instead.
Is TF the dominant tool in commercial or startup DNN projects?