In some datasets provided on kaggle, like the MNIST dataset, the data is in the form of .tfrecords which is a functionality of tensorflow from what I understand. While its useful, I’d like to know that if there is a way I can read it using keras or in fastai?
MNIST is such a standard data set that access to it is built in to many libraries, including pytorch. I recommend geting it from there rather than converting tfrecords.
To answer the question in the title, basically you use tensorflow to open the .tfrecords and write them out to another format, as exemplified here.