What is the dimension of image that is passed to the model for prediction ?

[lordzuko]

What is the dimension of image that is passed to the model for prediction ?

[zuphilip]

The input text lines can have different sizes but they are normalized before training and prediction, see also https://github.com/tmbdev/ocropy/blob/master/doc/line-normalization.ipynb such that the height of the text lines (the model will see) is fixed to 48 pixels AFAIK.

[lordzuko]

@zuphilip What about the width of image? Is there an upper limit which has been set, because AFAIK we can't pass variable dimension images to model, so what I can assume is since it is training on A4 size paper I can guess that the width of paper is the maximum width of image that is being input to the model? Please, Correct me if I am wrong.

[zuphilip]

Do you experience an error like "RecognitionError: input too large for LSTM model"? It seems that for the standard English model the N in https://github.com/tmbdev/ocropy/blob/f7b163c089d6db05a90bd52f6f60731ab4638b2d/ocrolib/lstm.py#L557 is set to 5000.