Dear authors,
I hope you are doing well. I am a student who has been studying your excellent work on the Text Glyph Structure Encoder. Thank you for making your research publicly available.
I have two questions:
When pre-training the Text Glyph Structure Encoder (as described in the paper), I noticed that only the training phase is performed, without separate validation or test stages. Could you kindly explain the rationale behind omitting validation and testing during pre-training?
I followed the configuration in GlyphTrain.yaml to conduct the pre-training. Among the generated checkpoints, would it be appropriate to select the one with the highest accuracy for use in subsequent UNet training?
Thank you very much for your time and assistance. I look forward to your guidance.
Dear authors,
I hope you are doing well. I am a student who has been studying your excellent work on the Text Glyph Structure Encoder. Thank you for making your research publicly available.
I have two questions:
When pre-training the Text Glyph Structure Encoder (as described in the paper), I noticed that only the training phase is performed, without separate validation or test stages. Could you kindly explain the rationale behind omitting validation and testing during pre-training?
I followed the configuration in GlyphTrain.yaml to conduct the pre-training. Among the generated checkpoints, would it be appropriate to select the one with the highest accuracy for use in subsequent UNet training?
Thank you very much for your time and assistance. I look forward to your guidance.