search

csguoh / LEMMA

[IJCAI2023] An official implement of the paper "Towards Robust Scene Text Image Super-resolution via Explicit Location Enhancement"
Apache License 2.0
52 stars 5 forks source link

pretrain_transformer.pth #23

Closed FocueAI closed 2 months ago

FocueAI commented 3 months ago

image Hi! I am very interested in this project. Can you tell me what project does the weight pretrain_transformer.pth come from?

FocueAI commented 3 months ago

Also, have you conducted an experiment using a Chinese character set? How should pre train_transformer. pth be replaced?

csguoh commented 3 months ago

Hi, thanks for your interest in this work. The pretrained_transformer is used for calculating the text position loss, which is proposed in this work. Since there is few work in Chinese text dataset, we are not able to conduct experiments on Chinese characters. If you want to train a Chinese character STISR models, you may need to re-train the pretrained_transformer for loss function, and also retrain the LEMMA model for Chinese character.

FocueAI commented 3 months ago

I really feel your answer! If I want to retrain the pretrained transformer, can you provide me with the corresponding training code? Looking forward to your answer and help again!

FocueAI commented 3 months ago

in addition in this work Corresponding code，I still can't find the training code corresponding to pretrained_transformer

csguoh commented 2 months ago

Hi, you may contact the author of STT for the training code of pretrained_transformer.pth.