We propose an efficient structure-aware labeling method to reconstruct both textual and tabular content in medical reports. It significantly reduces the human labeling errors and improves the efficiency of annotation process.
This system supports the integration of custom models for image processing and OCR. You can apply your own APIs in detector.py for the following optional and required models:
- Small angle detection model (optional)
- Large direction detection model (optional)
- Table angle detection model (optional)
- OCR model (required)
Once detector.py is configured, run run.py to get the ESRA result of an image