How to improve the recognition effect non-training set？

tomassosorio / OCR_tablenet

TableNet Implementation on Pytorch

144 stars 45 forks source link

How to improve the recognition effect non-training set？ #12

Closed Markson-Young closed 3 years ago

Markson-Young commented 3 years ago

Thanks for your project！ Table position detection performs well. But when I use non-training set images for testing ，The table information extraction effect is very poor ，using pytesseract OCR. Maybe there is a problem with my picture format?

tomassosorio commented 3 years ago

Hi @OY-JULIN1, it could be for multiple reasons, if for example, it's handwritten generally tesseract does not perform well, or, If you change the image aspect ratio it could also perform worse. Without more information, I am not able to help you...