Powerful handwritten text recognition. A simple-to-use, unofficial implementation of the paper "TrOCR: Transformer-based Optical Character Recognition with Pre-trained Models".
thanks for the great and simple repo!
Is the generation of the predict a greedy or a beam Search approach?
If you know how to implement an beam search generation, then I would be very happy if you could help me out!
Hi, I'm glad you enjoy using the repo! According to the original TrOCR paper, the Decoder already uses beam search to generate the output, see https://arxiv.org/pdf/2109.10282.pdf
Hi,
thanks for the great and simple repo! Is the generation of the predict a greedy or a beam Search approach? If you know how to implement an beam search generation, then I would be very happy if you could help me out!
Cheers, Jonas