3 years ago when I first got into deep learning, Tensorflow was definitely the dominated framework in Research. But now everyone around me are learning or already have been using PyTorch. I would say, first, PyTorch is much easier to get in hand especially for people familiar with Python. Also Tensowflow 2.X makes itself kind of dirty and messy. Lots of tf users prefer stick with Tf 1.X and reluctant to learn 2.X version. And PyTorch has an effective and active community which makes PyTorch more appealing.