feat(add): Handle airplane in the sky flying left challenge with SK-IMAGE solution

[QIN2DIM]

Introduction

Usage

You can run the demo project associated with the solution by running the following command:

# hcaptcha-challenge/src
python3 main.py demo_v3

Linked pull requests

@beiyuouo 30ae391

Demo

[adv-zl]

A lot of false-positive results occurred. How is it possible to train model or expand rainbow_table?

[QIN2DIM]

@adv-zl
The programme is designed with pass rates as an indicator, not accuracy.🛴 In terms of what you feel the programme's accuracy rate is probably at....

[QIN2DIM]

rainbow_table stores the MD5 value of the correct challenge image.Therefore, you need to do two things: collect enough correct challenge-images and convert them into MD5 strings.

[QIN2DIM]

SK-IMAGE is a technology for image recognition, this solution does not involve knowledge of deep learning.

Noteworthy, feature engineering can be extremely fast when set up properly, but it is a very difficult mathematical problem to improve the generalisation and accuracy of feature engineering.

[QIN2DIM]

Obviously, we will improve the solution and increase the accuracy rate in subsequent versions.

Of course, I would love to share my thoughts on the optimisation of the solution, but there are some obstacles to this. On the one hand, the optimisation problem is too abstract, and secondly, I am concerned that I am not able to express my ideas accurately in English.

[skrrppop]

Yeah i ran 100 it times and only got it 15 times working

[QIN2DIM]

@skrrppop WHAT👀 I don't quite understand what you mean.Do you mean calling methods in the lab environment to test on local images, or are you using the demo to register a bunch of bot accounts?

[QIN2DIM]

Yeah i ran 100 it times and only got it 15 times working

I'm trying to upgrade the solution, but I can't find a sitekey that consistently refreshes this label.

[QIN2DIM]

rainbow_table stores the MD5 value of the correct challenge image.Therefore, you need to do two things: collect enough correct challenge-images and convert them into MD5 strings.

It is worth stating that the rainbow_table containing the MD5 of correct images is not fully manually annotated. I will undertake a large scale challenge based on existing methods and filter outputs based on the annotation results given by the model.

[skrrppop]

Yeah i ran 100 it times and only got it 15 times working

I'm trying to upgrade the solution, but I can't find a sitekey that consistently refreshes this label.

Discord site_key="f5561ba9-8f1e-40ca-9b5b-a0b3f719ef34", site_url="https://discord.com",
Here are only airplane in the sky flying left

[skrrppop]

@skrrppop WHAT👀 I don't quite understand what you mean.Do you mean calling methods in the lab environment to test on local images, or are you using the demo to register a bunch of bot accounts?

I tested it with the pyhcaptcha library to solve it without need of selenium

[QIN2DIM]

This sounds really cool. Is this the library you are referring to? PyCAPTCHA

[skrrppop]

I Used this Here https://github.com/h0nde/py-hcaptcha

[QIN2DIM]

Discord site_key="f5561ba9-8f1e-40ca-9b5b-a0b3f719ef34", site_url="https://discord.com", Here are only airplane in the sky flying left

Great help!

I discovered this particular update rule yesterday, but now after more testing I am convinced that the update of the challenge labels is not completely random.

hcaptcha-challenger/src/services/settings.py

Line 15 in 3c40a94

"discord": "f5561ba9-8f1e-40ca-9b5b-a0b3f719ef34",

[adv-zl]

Yeah i ran 100 it times and only got it 15 times working

I'm trying to upgrade the solution, but I can't find a sitekey that consistently refreshes this label.

Better use sitekey=4c672d35-0701-42b2-88c3-78380b0db560

[QIN2DIM]

Thank you for your work.

Apparently there is a regular refresh of the challenge labels, airplane in the sky flying left only appears in some specific time period. This time period may be fixed, or it may be related to the heat of the site visit.