Questions about Fastai vs. TensorFlow, and Optical Character Recognition (OCR)

Today Jeremy has talked about the comparison and relationship between pytorch and fastai, I’m also curious on the perspectives that Fastai outperforms Tensorflow. Would you mind to briefly explain on that? @jeremy

Another question, in terms of Optical Character Recognition (OCR) what packages and functions would you recommend to use? A lot of blogs recommend Tesseract but I wonder whether Fastai or TensorFlow can also do well?

Thanks for your time! Any opinion or suggestion would be greatly appreciated!

In short - the kinds of tricks we showed on Monday night aren’t part of Tensorflow. They can, of course, be written in Tensorflow, just as we’ve written them in Pytorch - but there isn’t currently library like fastai for Tensorflow (and it would be quite a bit harder to create one, compared to using pytorch).

You can certainly do OCR with fastai or Tensorflow, although it’s easier to start with something that’s got the basics sorted out for you. Here’s a project that shows you how much is required to do OCR really well with deep learning: https://blogs.dropbox.com/tech/2017/04/creating-a-modern-ocr-pipeline-using-computer-vision-and-deep-learning/

2 Likes

Thank you very much for the help!

This may also help you https://hackernoon.com/latest-deep-learning-ocr-with-keras-and-supervisely-in-15-minutes-34aecd630ed8 :slight_smile: .

4 Likes

Thanks! Really appreciate it!

i find this list of OCR resources to be quite comprehensive and well maintained

3 Likes