On what scale is the text_score parameter from output of readtext()

[jkcg-learning]

What is the range of text_score parameter from output of readtext()

For some inference, I am seeing the confidence score of text above 1.00 or (100%)

[gaotongxiao]

It depends on the model that you've been running. We didn't implement a universal range limit on scores so it's only a relative confidence indicator to the same model for now.

[jkcg-learning]

I have used the following code for the inference on "demo_text_ocr.jpg" file
It is loading PANET(detection) and SEGOCR(Recognition).

from mmocr.utils.ocr import MMOCR
ocr = MMOCR()
results = ocr.readtext('demo/demo_text_ocr.jpg', print_result=True,details=True, imshow=False)

Output

{'filename': 'demo_text_ocr', 'result': [{'box': [218, 44, 157, 39, 160, 12, 220, 16], 'box_score': 0.9534332156181335, 'text': 'sale', 'text_score': 1.0}, {'box': [66, 60, 33, 60, 33, 43, 66, 43], 'box_score': 0.9697706699371338, 'text': 'sale', 'text_score': 0.9623287671232876}, {'box': [582, 117, 457, 109, 461, 55, 586, 63], 'box_score': 0.9692338109016418, 'text': 'all', 'text_score': 1.3333333333333333}, {'box': [706, 133, 580, 117, 587, 62, 713, 78], 'box_score': 0.9724510908126831, 'text': 'lyear', 'text_score': 1.0}, {'box': [83, 120, 21, 114, 23, 88, 85, 94], 'box_score': 0.975725531578064, 'text': 'sale', 'text_score': 1.0}, {'box': [663, 247, 447, 228, 452, 170, 669, 189], 'box_score': 0.9692399501800537, 'text': 'round', 'text_score': 1.0}, {'box': [514, 301, 462, 294, 464, 277, 516, 283], 'box_score': 0.95651775598526, 'text': 'aretal', 'text_score': 1.0}, {'box': [584, 309, 516, 302, 518, 283, 585, 290], 'box_score': 0.9467790722846985, 'text': 'loran0sg', 'text_score': 0.9693693693693693}, {'box': [584, 308, 584, 288, 637, 288, 637, 308], 'box_score': 0.9175753593444824, 'text': 'oveano', 'text_score': 1.1452991452991454}, {'box': [637, 308, 637, 292, 673, 292, 673, 308], 'box_score': 0.9137787818908691, 'text': 'ostcms', 'text_score': 1.0}, {'box': [496, 320, 449, 312, 452, 294, 500, 302], 'box_score': 0.9457231163978577, 'text': 'soceci', 'text_score': 1.0}, {'box': [587, 331, 499, 323, 501, 303, 588, 311], 'box_score': 0.9494416117668152, 'text': 'cnounsons', 'text_score': 0.9797979797979798}, {'box': [660, 328, 590, 328, 590, 312, 660, 312], 'box_score': 0.951109766960144, 'text': 'scronma', 'text_score': 0.9240506329113923}, {'box': [557, 415, 473, 404, 477, 373, 561, 383], 'box_score': 0.9712486267089844, 'text': 'ocbc', 'text_score': 1.75}, {'box': [560, 414, 560, 385, 620, 385, 620, 414], 'box_score': 0.9603750705718994, 'text': 'bank', 'text_score': 1.25}]}

box_score is this on a scale (0.00 to 1.00) or (0% to 100 %) ????

text_score scale is not much clear. In some cases it is exceeding above 1.00 or 100% .

What shall we conclude about the confidence score for box and text?