I am collecting images of Bus name boards for bus name board identification project that i am currently working on.
(Since the bus name boards are in regional languages(mostly) in my city, i do not want to do OCR. Instead i want to see if i can solve as an image classifier problem. ) I want to use CNN with fastai.
I have images taken under different conditions - sunny, cloudy, night, different angles. I have images like
Would it be better to take the portion of image where the name board alone is there and create images like these ?
In this case, the images are not going be to of the same size. will that be an issue?
Any help would be appreciated.