Hi folks,
Last week Tulasi and I
wrote to you regarding our work on creating text detectors in TensorFlow Lite. Tulasi extended that work and developed OCR models in TensorFlow Lite. Read further if this sounds interesting.
Folks working in the area of OCR might already be familiar with the library called
Keras OCR. He took the pre-trained models provided by them and successfully converted them to TensorFlow Lite. To my knowledge, he is probably the first one that has been able to do it. The converted TensorFlow Lite models are to be published on TensorFlow Hub (a
PR is on the way).
TensorFlow Lite team, hear hear!
Mobile application developers, this might another excellent opportunity for you to create a sample application out of these models for demonstration purposes.
Extensions:
We now plan to extend this to support the use of machine translation and text-to-speech considering their real-world impact. If you are interested in collaborating please follow the details
here and express your interest.