Datasets?

[sagarhukkire]

Hi,

It is possible for you to share labels and image files(.npy) and trained classifier on NIST files. It will be a great help for me. Currently, I figured out how to get the character out of the word, but CPU is not helping me train on the entire dataset.

Thanks in advance.

Sagar

[sagarhukkire]

did it finally

[garchit33]

@sagarhukkire can you please tell me how do you segment charcater out of word.
I will be very thankful to you.

[sagarhukkire]

Hi First thing you do is define Region of Intrest for the fields on Forms. 1. ROI contains only handwritten words 2. Apply connected components (8 or 4) , then you will get bounding box for each character, crop each one store 5) Train CNN on NIST database, they have million of handwritten characters images 6) now this CNN apply to your cropped images ...for every crop character you will get prediction and you bind them For example 1. ROI is : USA 2. Connected components : U, S, A 3) CNN predictions : USA ( after binding ) To improve you can put SVM at last layer of CNN Hope it will help you Thanks and Regards Sagar Hukkire

…

On Aug 15, 2018 9:51 PM, "garchit33" ***@***.***> wrote: @sagarhukkire <https://github.com/sagarhukkire> can you please tell me how do you segment charcater out of word. I will be very thankful to you. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#5 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ATFxy6PmEwkrSJWVLo9QoQSra_S1aIIOks5uRHuwgaJpZM4T7Uy8> .

[garchit33]

Do you have any implemented approach of this. Can you please provide me with the code of this if you have. Regards Archit Garg On Thu, 16 Aug 2018, 11:10 am Sagar Hukkire, <notifications@github.com> wrote:

…

Hi First thing you do is define Region of Intrest for the fields on Forms. 1. ROI contains only handwritten words 2. Apply connected components (8 or 4) , then you will get bounding box for each character, crop each one store 5) Train CNN on NIST database, they have million of handwritten characters images 6) now this CNN apply to your cropped images ...for every crop character you will get prediction and you bind them For example 1. ROI is : USA 2. Connected components : U, S, A 3) CNN predictions : USA ( after binding ) To improve you can put SVM at last layer of CNN Hope it will help you Thanks and Regards Sagar Hukkire On Aug 15, 2018 9:51 PM, "garchit33" ***@***.***> wrote: > @sagarhukkire <https://github.com/sagarhukkire> can you please tell me > how do you segment charcater out of word. > I will be very thankful to you. > > — > You are receiving this because you were mentioned. > Reply to this email directly, view it on GitHub > < #5 (comment) >, > or mute the thread > < https://github.com/notifications/unsubscribe-auth/ATFxy6PmEwkrSJWVLo9QoQSra_S1aIIOks5uRHuwgaJpZM4T7Uy8 > > . > — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#5 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/Ac3VinjAukmR9dkY4X8bo9g4_vOMj1-fks5uRQXggaJpZM4T7Uy8> .

[sanksys]

@sagarhukkire
hi sagar could you please provide the code (GitRepo link) for comment 6 implementation
mostly for
2. Apply connected components (8 or 4) , then you will get bounding box
how you approached this ?

thanks
Sanket
Sankaghav123@gmail.com