THIS REPO IS UNDER DEVELOPMENT
This repo provide end-to-end CNNs for text detection and recognition based on RetinaNet. Currently, two solutions are proposed:
Facebook's Rosetta system for text detection and recognition in images (source: Facebook [3])
One-stage architecture for text detection and recognition in images. Image is adapted from Rosetta's paper (Sorry I'm not good at drawing)
[1] RetinaNet: https://arxiv.org/abs/1708.02002
[2] CTC loss: http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.139.5852
This repo was built using several materials as below.
Keras RetinaNet: https://github.com/fizyr/keras-retinanet
Keras OCR: https://github.com/keras-team/keras/blob/master/examples/image_ocr.py