What problem does it solve?
Extracts line-level and full recognized text from images and PDFs when you need exact text content and bounding-box locations rather than high-level layout or table reconstruction. It addresses the common need to convert screenshots, photos, scans and single-page PDFs into editable, machine-readable text with positional data.
Core Features & Use Cases
- Line- and box-level OCR: Returns complete recognized text plus line-level bounding boxes suitable for annotation, search, or downstream parsing.
- Multiple inputs: Accepts file URLs, local file paths, or uploaded images (base64 payloads) and auto-detects image vs PDF where possible.
- Robust output handling: Saves a stable JSON envelope containing a complete text field and raw provider result, and prints the saved path by default.
- Operational tooling: Includes a smoke test to validate configuration and a CLI caller script for production use.
- Use case: Extract text lines and coordinates from scanned invoices, screenshots, or document photos for data extraction or archival.
Quick Start
Run the ocr_caller script with a file URL or local file path after configuring PADDLEOCR_OCR_API_URL and PADDLEOCR_ACCESS_TOKEN.
