Hi there,
I think there are few issues with vertical words.
Shouldn't this function https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/easyocr.py#L344 output max_width as well in order to update max_width = max(max_width, imgH) in the next line ? It seems that if there is a long vertical word then it's capped by imgH and recognition is usually wrong.
Also, I realized that images are cropped and resized in here https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/easyocr.py#L341 based on their ratio which makes long image crops, that is h >> w, very small (their width is squeezed a lot). Then, these resized images are rotated (90, 180 and 270) in https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/easyocr.py#L344. I think the images should be rotated before they get resized.
Hi there,
I think there are few issues with vertical words.
Shouldn't this function https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/easyocr.py#L344 output
max_widthas well in order to updatemax_width = max(max_width, imgH)in the next line ? It seems that if there is a long vertical word then it's capped by imgH and recognition is usually wrong.Also, I realized that images are cropped and resized in here https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/easyocr.py#L341 based on their ratio which makes long image crops, that is h >> w, very small (their width is squeezed a lot). Then, these resized images are rotated (90, 180 and 270) in https://github.com/JaidedAI/EasyOCR/blob/master/easyocr/easyocr.py#L344. I think the images should be rotated before they get resized.