2nd can be filled with OCR

2nd can be filled with OCR

I tried using pytesseract (code), it gives rather unsatisfactory results. I tested it for 3 captchas, it gave correct for only 1. A neural net trained specifically for the given dataset would surely outperform pretrained OCR.
I have obtained the dataset from aims using a script to download 30k captcha images from AIMS and then separated them in form of individual characters, the dataset can be found here. However, I have not started work on CNN as of yet.

@RachitKeertiDas

Hi,

So, an ideal solution would definitely be filling both endpoints.
Also, the PR introduces content_scripts. This would also require us to have additional justification while uploading our extension, and additional review time.
Another small nitpick is that once a captcha expires, we need to refill it. But that can be handled later.
There are also possibly some administrative issues in adding this functionality officially, but I think they can be worked around.

For filling the second captcha, we can see later. Ideally, we would like all our code to be bundled together, and not call an external server to just infer the captcha from the image for security reasons. So, you might have to look into solutions such as tensorflow.js to allow inference in js in the bundled extension itself.

Good work on obtaining the dataset though.

P.S. I'm not really an active maintainer for this repository, so my suggestionss might not be of much value in getting the PR Merged.