Transfer learning: text classification on images

I have a image classification problem that is essentially classifying images of pdf documents into categories. I want to use transfer learning to make use of an existing network architecture for text in images. It feels like using something like VGG or Inception would be overkill because the feature space of our problem is constrained images of pdf files.

I am wondering if there are any well know architecture/weights for text recognition in images that would be particularly well suited for this problem?

ICDAR holds annual competitions on text recognition - have a look at previous solutions.

1 Like