search

clovaai / donut

Official Implementation of OCR-free Document Understanding Transformer (Donut) and Synthetic Document Generator (SynthDoG), ECCV 2022
https://arxiv.org/abs/2111.15664
MIT License
5.71k stars 462 forks source link

Non-square images ? #121

Open MohamedAliRashad opened 1 year ago

MohamedAliRashad commented 1 year ago

How to train the model on a non-square images ? I tried an input of 2200x1700 and this is the error i got:

RuntimeError: shape '[1, 55, 10, 42, 10, 1]' is invalid for input of size 233750
Wyzix33 commented 1 year ago

It seems like input needs to be multiple of 320...