A Chrome and Firefox browser extension containing the EfficientNet neural network. The extension automatically detects reCAPTCHA on a page, clicks the checkbox, recognizes the task, solves it with the neural network, and clicks the correct squares. Then it clicks “verify,” and if a new task appears, it solves it until the green checkmark shows up. Besides the neural network, the extension also includes an SQLite database. Each image, before being sent to the neural network, is checked against this database. Currently, the database only contains images that have already been encountered and where the neural network made mistakes, reducing the number of AI errors.
It’s possible if there are many requests for it, but it’s not a priority right now. Tests show that the neural network consumes very few CPU resources, even much less than the browser uses to render reCAPTCHA.
A pHash is calculated for every image. If this pHash has already appeared and was marked as incorrect, it is stored in the local SQLite database — the extension checks the local database before sending the image to the neural network, and if it’s found there, it takes the solution from the database. This way, if the image was encountered before and the neural network got it wrong, the local database corrects the error.
That’s in the plans, but for the long term. We also need to see what’s cheaper for the CPU: searching a large pHash database or running neural network recognition.
Yes.
Not yet. I was specifically interested in beating reCAPTCHA. Maybe in a paid version, if there’s enough demand.