Authors: Shimaa Elabd,m
Description: Make everything simple, solve the most complex problems by simplifying solutions.
Everything depends on the limits of trust. If you go trust? Everything went.
But surrender is synonymous with nothingness. We must never stop.
The idea is to cut the image so that we can get every part of the picture. Then ensure that it is properly sized.
So we can feed it to a strong classification network, so that it is able to see some of the features of the cropped part ..
Another step is to find the characters in those areas, while being careful to detect the rotation.
Anchors are not useful in many cases with different text sizes and rotation.
Using machine learning tools in image shredding such as SVM or KNN instead of anchors is sometimes more beneficial. ...As an Assistant factor ???
Maybe the idea is not logical but what I can say about how to work. It is a complete building that includes all the steps end to end.
Not all illogical ideas are wrong. Some are great
But he needs faith in him and sacrifice for making it true.
Finally we added the possibility of detecting the round text and achieved a yield of more than 0.96
"2019-01-22 fast_ret_sh_05 95.19% 97.92% 96.53%"
But it did not seem to be useful to add.
The idea is mainly for the researcher Shimaa Elabd.