How 81newark/anticap works
Project type and purpose
• AntiCAP is a Python library/tool (not a full-stack app) for recognizing various CAPTCHA types:
• OCR text CAPTCHAs (via DdddOCR)
• Arithmetic CAPTCHAs (via a PyTorch model)
• Sliding‐puzzle CAPTCHAs (gap‐match and shadow‐comparison)
• It exposes a simple Python API (with future plans for a WebApi) and relies on pre‐trained ML/ONNX models.
File structure
Top level
• setup.py, requirements.txt, main.py – installation, deps, CLI entrypoint
• README.md, LICENSE
• AntiCAP/ (package)
– __init__.py: defines class AntiCAP() and dispatch methods
– Arithmetic.pt: PyTorch model for arithmetic CAPTCHA
– Ddddocr.onnx: ONNX model for text OCR
– README.md, requirements.txt: module‐specific docs & deps
No “frontend” or “backend” directories: this is a library/service component.
Key insights from README
• Describes supported CAPTCHA types and usage examples
• Mentions future Web API layer
• Points to a separate “AntiCAP_trainer” repo for training new models → extensibility
Components & interactions
a. Main components
1. Client (Python script or future HTTP client)
2. AntiCAP Controller (AntiCAP.AntiCAP class)
3. Strategy Modules:
• DdddOCR Module (loads Ddddocr.onnx via onnxruntime)
• Arithmetic Module (loads Arithmetic.pt via PyTorch)
• Slide-Match Module (image processing logic)
• Slide-Comparison Module (image comparison logic)
4. Model Artifacts
• Ddddocr.onnx
• Arithmetic.pt
5. Dependencies & Runtimes
• Python 3.8
• PyTorch
• onnxruntime
• OpenCV / PIL for image decoding
6. (Optional future) WebApi Layer
7. (External) AntiCAP_trainer service
b. Relationships & data flow
1. Client → AntiCAP Controller (calls specific method, e.g., .Arithmetic(img_base64))
2. Controller decodes Base64 → dispatches to correct Module
3. Module loads corresponding model or uses comparison algorithm
4. Module returns result to Controller → back to Client
5. (Future) Controller may also expose HTTP endpoints
c. Architectural patterns & design principles
• Strategy pattern: each CAPTCHA type is a pluggable “strategy module” under a unified API
• Dependency inversion: high-level Controller depends on abstract module interfaces; modules depend on ML runtimes
• Separation of concerns: image I/O, model inference, result formatting are in separate layers
d. Significant technologies & libraries
• Python 3.8, pip package
• PyTorch for .pt models
• ONNX Runtime for .onnx models
• OpenCV / PIL for Base64 image decoding
• Future: FastAPI or Flask for WebApi