search

mjq11302010044 / TATT

A Text Attention Network for Spatial Deformation Robust Scene Text Image Super-resolution (CVPR2022)
MIT License
164 stars 17 forks source link

Is the OCR evaluation model (ster\crnn) and tatt end-to-end? #8

Open HansonnnCheung opened 2 years ago

HansonnnCheung commented 2 years ago

Is the OCR evaluation model (ster\crnn) and tatt end-to-end ? OR first use SR model to output results, and then input OCR? just like the code below： def getitem(self, index): ... ... label_str = str_filt(word, self.voc_type) return img_HR, img_lr, img_HRy, img_lry, label_str

Does “label_str” participate in the training of the whole model？

mjq11302010044 commented 2 years ago

@HansonnnCheung Not really, TATT follows the settings in the TPGSR, i.e., the label_str is not used in training.

HansonnnCheung commented 2 years ago

Thank you for your answer. Where did the “teaching_loss” come from during the training. Is "TEXT_Prior" generated by an OCR model? @mjq11302010044

In addition, if "lable_str" does not participate in training, is it necessary to use the value of AAA as an input during DataLoader?

HansonnnCheung commented 2 years ago

Do you mean that “TPGSR and its settings” is the TPGSR model，which published in your home page? @mjq11302010044