Paddle OCR Detection confidence level

Question

I have a question regarding Paddle OCR detection model. Paddle OCR output results are the coordinates of the bounding boxes, recognized text and the confidence score, and from what I have researched online the confidence score here is regarding to the recognizer. is there a way to obtain the confidence score of the detected text? In my situation I would like to know how confident Paddle OCR is in detecting text and how sure is it that the thing that it detects is a text and not just random scribbles or just rust on a shipping container?

Thanks in advance!

Here is my ocr result:

OCR Result: [[[[[641.0, 65.0], [813.0, 61.0], [815.0, 130.0], [643.0, 134.0]], ('FRLU', 0.9880595207214355)], [[[645.0, 156.0], [953.0, 152.0], [954.0, 214.0], [645.0, 217.0]], ('8616911', 0.9634677767753601)], [[[640.0, 243.0], [816.0, 243.0], [816.0, 307.0], [640.0, 307.0]], ('42G1', 0.9629315733909607)]]]

Answer 1

To check the confidence score for each individual character recognized by PaddleOCR, you can modify the decode() function in the BaseRecLabelDecode class. This class is located in your virtual environment at:

venv/lib/python3.9/site-packages/paddleocr/ppocr/postprocess/rec_postprocess.py

By default, the OCR returns the mean confidence score for all characters in the detected text within a bounding box. Updating the decode() function will allow you to access the confidence score of each character individually after the recognition process is completed.