Thanks for your great work!
I want to train my own dataset, however the image are not the same size, curent method will convert them all to 512x512(seems using crop), will this affect the training performance, Is there a proper way to train these images?
I haven't seen a config to do this yet, but it should be possible. You'd need to alter the data pre-processing for train and val datasets, but it should work.
Thanks for your great work! I want to train my own dataset, however the image are not the same size, curent method will convert them all to
512x512
(seems using crop), will this affect the training performance, Is there a proper way to train these images?