GSA Ser users - could I ask a favour ?

[ATuringtest]

I'm working on a machine learning image captcha breaker (wrapped around TensorFlow) that can run locally (if you have a decent GPU).
This is the first stage - simple images - numerical and alphabetical etc.
What I would like if you can - could you please save your solved captchas in GSA Ser.
There is a bit of a hitch here though - GSA captcha breaker reports quite a lot of unsolved captcha's as solved which is no good to me.
I've been spending few dollars to use 2captcha for image captchas because I know the results are 100%.
So if you happen to be using a human captcha service and could send me the solved ones I would be grateful
Just PM me with an attachment of the solved ones zipped etc or if its to big I'll give you and email to send it to that's 50 meg plus
So far I have a mix of images solved of about 10k but its not really enough I would like more of other types and about 10k of each of those types to really make it 100% solved ( fingers crossed)
typical would look like this with a file name of the actual correct solved captcha

|| example : 409036.jpg



Cheers

 

[JasonS]

I use CapMonster Pro. If that would help, then let me know I'll enable the save captcha option there. Before I was using captcha.guru service and spent a few hundred bucks over there.

 

[ATuringtest]

I use CapMonster Pro. If that would help, then let me know I'll enable the save captcha option there. Before I was using captcha.guru service and spent a few hundred bucks over there.

Yes please mate that would be great

 

[JasonS]

Yes please mate that would be great

After enabling this setting, it saves the reCAPTCHA images as well. Which is just taking the server storage resources. Not sure where to just set it to text image captchas. I'll ask their support about this.

 

[ATuringtest]

After enabling this setting, it saves the reCAPTCHA images as well. Which is just taking the server storage resources. Not sure where to just set it to text image captchas. I'll ask their support about this.

thanks all help appreciated

 

[TZ2011]

For correct recognition you will need millions of these images and still not be sure for final results. even human sometimes don't have a clue is it O or 0, 6 or b, 1 or l or I, B or 8 ... Isn't it easier to train the OCR model, which is specifically designed for visual recognition issues?

 

[ATuringtest]

For correct recognition you will need millions of these images and still not be sure for final results. even human sometimes don't have a clue is it O or 0, 6 or b, 1 or l or I, B or 8 ... Isn't it easier to train the OCR model, which is specifically designed for visual recognition issues?

It is OCR but using an AI model to process it more accurately - quickly and also reinforcement by reward training which cannot be taught to a standard program.
And it does not take millions of images for a data set ..
Question : "How can we perform efficient inference and learning in directed probabilistic models, in the presence of continuous latent variables with intractable posterior distributions?"
Answer (possibly) = Auto-Encoding Variational Bayes
I'm either having major da ja vu here or a brain aneurism lol - I'm sure I answered this same question in another thread ...?

 

[WhiteHatCoder]

Huh learning AI myself, this seems interesting.
Can you please share current results of testing data set in terms of efficiency in %.

This is going to be nice service here on BHW

 

[ATuringtest]

Huh learning AI myself, this seems interesting.
Can you please share current results of testing data set in terms of efficiency in %.

This is going to be nice service here on BHW

Early days and its not my primary project at the moment. But I had it at 98% with a single captcha numeric type. Hence why I'm trying to get more types.
Also the reinforcement training means that the more it attempts to solve.. the greater % of correct solved captcha's increases.