This network is used for optical character recognition which comprises of text detection and text recognition. The network is composed of a ResNet-FPN feature extractor, a detection branch, and a recognition branch. The model is trained by the ICDAR-2017 dataset. The input is an image containing some character. The output is a structure that includes the words recognized and their position. The following image shows the result of OCR.
The following table lists the OCR models supported by the Vitis AI Library.