How hamster-yang/ai_captcha works
Main Components
A. Data Processing Pipeline
• Input Sources (marked in blue)
• Video Capture System (based on video-related files)
• Static CAPTCHA Images
• Training Data Generation
B. Model Training System (marked in green)
• Multiple Model Versions (evident from model_*.h5 files)
• Training Scripts
• Model Evaluation
C. CAPTCHA Recognition System (marked in orange)
• Image Preprocessing
• Character Segmentation
• Recognition Engine
• Output Processing
[Data Collection Layer]
• Video Input ──────┐
• Static Images ────┤
• Generated Data ───┘
• ▼
[Preprocessing Layer]
• Image Processing
• Data Augmentation
• ▼
[Training System]
• Model Architecture
• Training Pipeline
• Model Versioning
• ▼
[Recognition System]
• Image Capture
• Preprocessing
• Character Recognition
• ▼
[Application Layer]
• Web Integration
Key Elements to Highlight
a. Data Flow Directions:
b. Component Categories:
• Processing Units (rectangles)
• Input/Output Points (rounded rectangles)
• Decision Points (diamonds)
Notes for Implementation
• Clear separation between training and inference paths
• Data validation points
• Model checkpoint locations
• Error handling paths