OCR-CAPTCHA
| Property | Value |
|---|---|
| Author | xiaolv |
| Model URL | https://huggingface.co/xiaolv/ocr-captcha |
| Base Model | DAMO Text Recognition Model |
What is ocr-captcha?
OCR-CAPTCHA is a specialized optical character recognition model designed specifically for identifying and decoding various types of CAPTCHA verification codes. The model comes in two variants: a small model trained on 84,000 CAPTCHA images achieving nearly 100% accuracy, and a larger model trained on 1.35 million images with 93.95% accuracy.
Implementation Details
The model is fine-tuned on DAMO Academy's text recognition base model, specifically optimized for handling CAPTCHA recognition tasks. It supports multiple CAPTCHA formats including numeric-only, alphanumeric, and pure alphabetic (both uppercase and lowercase) combinations, with sequence lengths of 4, 5, or 6 characters.
- Small model: 700MB dataset, 27 training epochs, ~100% accuracy
- Large model: 11GB dataset, 1 training epoch, 93.95% accuracy
- Supports multiple CAPTCHA types and lengths
Core Capabilities
- Recognition of pure numeric CAPTCHAs
- Processing of mixed alphanumeric codes
- Handling of pure alphabetic sequences (case-sensitive)
- Support for variable length sequences (4-6 characters)
- High accuracy recognition across different CAPTCHA styles
Frequently Asked Questions
Q: What makes this model unique?
The model's specialization in CAPTCHA recognition and its high accuracy rates, particularly the small model's near-perfect performance, make it stand out. It's specifically optimized for common CAPTCHA formats found in real-world applications.
Q: What are the recommended use cases?
This model is ideal for automated CAPTCHA recognition in testing environments, web automation tasks, and accessibility tools. The small model is recommended for most use cases due to its superior accuracy and efficiency.
