Its pretty hard to nail them down manually - or maybe Im doing it wrong.
That's not all problematic captchas -there's quite a bunch of them actually. My wild guess is that CB somehow misidentifies those captchas. - example http://awesomescreenshot.com/0282uuum6e
I have 30 percent threshold and no tick in Try to Solve so my solve ratio should be pretty high, whilst its not. That how I noticed this strange thing so it can possibly be also replicated like that?
hmm I would need the URLs where that was coming from. Most of those captchas are new types that got identified wrong. I need the URLs to download more samples and add them.
Well it would take many hours to dissect them manually because log is very limited in size and captchas - engines often mixed up.
I sent you the lists with targets that contain many of problematic captchas and project with the engines that will catch many of said captchas - maybe you could debug it more easily, because with log babysitting it could take really long time.
just save unsolved captchas and send them to me. The domain is in the filename so I then know where to find more capthas of that type and can add it then.
Unfortunately its nigh impossible for me to filter low probability captchas (<50, considering that we want to find those with high solve rations that were unsolved) from this mix..
Comments