VKCaptchaSolver

Use this VKontakte captcha breaker with 75-84% accuracy with Python, JS, 2023.

It works with RU or ENG captchas up to 7 characters long.

Steps to start CaptchaSolver:

1. Clone the repo

git clone https://github.com/Venom998/VKCaptchaSolver.git

2. cd captcha_solver
3. Install all requirements: npm install
4. Run node start_node.js

Prerequisites

Change ai_model_lang_eng = false to true in start_node.js if you want to use captchas with English characters only.

model_eng.onnx - supports only "English" captchas.

model_ru.onnx - supports only "Russian" captchas.

Steps to train AI model:

0. Install all requirements(if not yet) using command pip install -r requirements.txt
1. download images from here (or use your own dataset), create folder images and place all dataset into 2 folders: train and test. Image name format: answer123-randomseed.jpeg
2. Configure code/utils/config.py file - img_width, img_height, format, ect.
3. Run code/statistics.py for analyzing your dataset (it will print all characters, whose your captchas contains)
4. code/main.py - TRAIN your model
5. code/test_mode.py - Test your tensorflow model
6. code/to_onnx.py - Convert your model to ONNX format
7. code/onnx_tester.py - Test your ONNX model
8. Write your own captcha-solver-adapter. For example you can look at this file

code/tools.py - конфиг файл (файл модели, ) + тулзы

code/test_mode.py - тест модели на реальных данных code/to_onnx.py - конверт модели tesorflow в формат onnx (для использования через onnxruntime)

vk_captcha_lib - код библиотеки vk_captcha по решению капчи