Current Capabilities

Checkbox and Associated Text Detection

The Document Understanding Subnet currently supports advanced Checkbox and Associated Text Detection (available on Testnet 236). This feature accurately identifies checkboxes within document images and associates each with its related text.

It leverages a sophisticated vision model for checkbox location and an OCR engine to extract corresponding text. The high accuracy checkboxes detector of this feature surpasses that of leading centralized solutions, including GPT-4 Vision and Azure Form Recognizer, making it invaluable for processing structured forms and surveys. This detection capability significantly improves data extraction quality and efficiency, proving essential for applications that rely on structured form data.

Figure 2: Example of checkbox and associated text detection.

Figure 3: Example of checked text without boxes.