see icon indicating copy to clipboard operation
see copied to clipboard

Published 20 hours ago • verified publisher icon Bartzi

saved model

Open un-lock-me opened this issue 5 years ago • 10 comments

Thanks for sharing your code with us. Could you also share your saved model?

un-lock-me avatar Feb 20 '19 18:02 un-lock-me

I already shared the model, please have a look at this website, and also read the correspinding section of the README :wink:

Bartzi avatar Feb 21 '19 10:02 Bartzi

@Bartzi Can we build model for CAPITAL ALPHANUMERIC Character string recognition ?

harshalcse avatar Mar 01 '19 06:03 harshalcse

Sure :wink: Just find the necesary training data and use create a new char_map only with the ascii codes for captial alphanumerics. Train and enjoy, yhis should be it...

Bartzi avatar Mar 01 '19 09:03 Bartzi

@Bartzi so what is preriqusite of dataset preparation like image resolution and image size and image quality...

harshalcse avatar Mar 06 '19 06:03 harshalcse

Hi,

good question. I never really cared about image resolution or quality, I mean the text should be readable, but that is about it. All samples that you put into the network will be resized to200x64 pixels (see here), regardless of their original aspect ratio.

So, you will need samples that only contain the already cropped text line and a ground truth file together with the samples. You can find a sample ground truth here.

Bartzi avatar Mar 06 '19 09:03 Bartzi

@Bartzi Is it required to create Dataset with bounding box file that is digitStruct.mat as it is there is SVHN dataset and if it is yes that how we can generate it ?

Please help.

harshalcse avatar Mar 07 '19 11:03 harshalcse

No, there is no need for bounding boxes! This is the point of the paper, you only need an image and the textual content as annotation. The README and the website I pointed you to should contai all information you need.

Bartzi avatar Mar 07 '19 12:03 Bartzi

No, there is no need for bounding boxes! This is the point of the paper, you only need an image and the textual content as annotation. The README and the website I pointed you to should contai all information you need.

@Bartzi Sir, I am asking for retraining and fine-tuning. Please help.

harshalcse avatar Mar 07 '19 12:03 harshalcse

I'm sorry but I don't see your problem. If you want to retrain, or finetune, you just have to prepare everything as if you'd like to train the model from scratch, but you can just use an earlier snapshot as a starting point.

Bartzi avatar Mar 07 '19 13:03 Bartzi

@Bartzi Sir the trained model provided is trained on which dataset please suggest and also please suggest using which .py file that model is trained.

harshalcse avatar Mar 09 '19 05:03 harshalcse