training on custom dataset

[azamshoaib]

@meijieru Thanks for this package. I would like to know that I want to train it on a different dataset, but when I use /tools/create_dataset.py, I get nothing. So please help me in this matter or suggest me any other method to do so.

[meijieru]

Could you please provide more information for analyzing your problem? Thanks

[azamshoaib]

@meijieru Thanks for your reply. I want to train the CRNN on my own dataset of korean characters so can you please guide me how to make dataset which can be used for training the CRNN.

[meijieru]

Do you follow the instruction described in the origin crnn?

[azamshoaib]

@meijieru As they gave a little instruction and i am a little novice so can you please help me. Thanks

[meijieru]

Does the creation success? Or are there any error message?

[azamshoaib]

@meijieru I still got error messages. "UnicodeDecodeError: 'utf8' codec can't decode byte 0x89 in position 0: invalid start byte" Don't know how to solve it. Please help me in this regard. Thanks

[meijieru]

This seems to be related to python version, maybe you could try python2 instead

[azamshoaib]

@meijieru Thanks for your reply. Can you please guide me how to train it using korean characters. I am using python2.

[meijieru]

This post may help you

[azamshoaib]

@meijieru thanks for your reply. I just want to know can I train it for Korean characters. Can you please give me some insight on this. I am really stuck at the point. Thanks

[meijieru]

It should work for thing like image->sequence problem, of course Korean characters.

[azamshoaib]

@meijieru
Thanks for your prompt reply. So can you please provide any tutorial to create dataset other than English characters and also instructions how to train it. This would be a huge help. Thanks. I tried my best but all in vain. My all images are of the same dimension(width and height) and all have the same length of characters in it.

[BigBorg]

Even for english characters, the one-line instruction is totally not enough.