method: CRNN after UNet Segmentation2019-04-22

Authors: Roberto Lotufo, Ramon Pires, Israel Campiotti, Rubens Machado, Luis Serrano, Giovanni Garuffi

Description: Word Recognition with an end-to-end neural network involving convolutional and recurrent stages. The recurrent stage encompasses two bi-LSTMs. The input of the current approach are cropped image patches gathered in the Task 1, based on DynamicUnet with two segmentations maps (the bounding box and the marker for the bounding box) and watershed pos-processing from markers to bounding boxes to disconnect vertical overlapping.