You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
(A dataset is being built, gt on the left, pred on the right)I want to recognize "월 간 야 수" properly, but when I grab the vertical area in CRAFT, the recognition does not work properly.
The issue I've read is [https://github.com//pull/450] and [https://github.com//issues/170].
However, in first case, character recognition is based on rotation according to the aspect ratio, and when the craft area is held vertically, the characters are not recognized properly. In second case, The vertical characters in my image are not separated properly.
I understand that learning will not work properly even if the vertical area is labeled. Is this correct? (I'm building a dataset, and I'm asking for advice because I think the workload will increase if the rules change.) Please give me advice on how to improve performance for vertical characters.
The text was updated successfully, but these errors were encountered:
Additionally, although learning is focused on Korean, when learning is conducted by adding uppercase and lowercase alphabets to characters, there is a problem that in the inference results, A is recognized as only lowercase letters, such as a. If anyone knows the cause, please respond.
(A dataset is being built, gt on the left, pred on the right)I want to recognize "월 간 야 수" properly, but when I grab the vertical area in CRAFT, the recognition does not work properly.
The issue I've read is [https://github.com//pull/450] and [https://github.com//issues/170].
However, in first case, character recognition is based on rotation according to the aspect ratio, and when the craft area is held vertically, the characters are not recognized properly. In second case, The vertical characters in my image are not separated properly.
I understand that learning will not work properly even if the vertical area is labeled. Is this correct? (I'm building a dataset, and I'm asking for advice because I think the workload will increase if the rules change.) Please give me advice on how to improve performance for vertical characters.
The text was updated successfully, but these errors were encountered: