Methodology for training

[mariembenslama]

Hello, Sorry for asking so many questions 😅 I was thinking of a way so that the program can grasp all the characters.

What if we start by teaching it for example a small amount of characters and then add the others little by little?

For example we give it a dataset of 50 characters only, we see how it performs, and then next time we add a dataset of new ones and see if it's able to differentiate the features.

Do you think an approach like this is fine? Or it's better to give it everything from the begining?

[niddal-imam]

Hi,

I have read an article mentioned an algorithm for training text recognition model called curriculum learning. I think it matches your idea. This is the title of the paper: Rosetta: Large Scale System for Text Detection and Recognition in Images.

I hope it helps.

[mariembenslama]

Thanks for the answer, I just checked out the articles and it seems interesting! But I wonder if it will work in our case. Feeding the algorithm little by little will that really bring high accuracy?

But I guess going with Holmeyoung's idea about large dataset = better answer in this case is probably the right thing.

But due to my machine's limits and since I'm using google colab (12 hours free gpu use only) as a tool for DL is giving me a disadvantage here when creating a large dataset 😌 That's why I thought about this idea.

Anyways, thanks a lot for the answer 😄 Also do you recommend any DL tools/environnement?

[niddal-imam]

No problems. Actually, I am using paperspace, but it is not free unfortunately 😌.

[mariembenslama]

Hehe I see,it's troubling right 😅

Well, it's worth it when your program work 😊

Thanks for the suggestion and the answer :)

[niddal-imam]

Yes, it is worth it.

Goodluck.