How nladuo/captcha-break works
Project type and purpose
• This is an open-source CAPTCHA-breaking toolkit (“captcha-break”) that provides end-to-end pipelines—implemented in both C++ and Python—for downloading, preprocessing (splitting), recognizing, and (in some cases) training models on CAPTCHAs from multiple sources (basic, CSDN, SubMail, Weibo.cn, JiKeXueYuan, Weibo.com).
• It uses OpenCV for image processing, Tesseract (or a custom tiny-cnn library) for OCR, and includes data-generation/training modules for model refinement.
File structure
• Top level: LICENSE, README.md, and six source modules: basic, csdn, submail, weibo.cn, jikexueyuan, weibo.com
• Each module (e.g. csdn) is subdivided by language:
• cpp/… with CMakeLists.txt, downloader, spliter, recognizer subfolders
• python/… with downloader, spliter, recognizer scripts + requirements.txt
• The “weibo.cn” and “weibo.com” modules also include a trainer subfolder, complete with tiny-cnn code (in cpp) or Python training scripts.
• Config/build artifacts: CMakeLists.txt, requirements.txt, .gitignore files, data archives (dataset.zip, tests.zip) and weight files.
README for additional insights
• Lists each supported CAPTCHA source, with screenshots.
• Emphasizes a uniform three-stage pipeline (download → split → recognize) plus optional training.
Detail
• Six source modules as separate “pipelines” or group them under a “multi‐source” umbrella, each following the same internal flow
• Where datasets (e.g., dataset.zip, tests.zip, captchas_backup.zip) and model-weights reside
• Build/deploy elements: CMake for C++, requirements.txt for Python