A confidence value for evaluation?

[shin7]

Thank for your great work!

for bid, box in enumerate(detections):

    boxo = detectionso[bid]
    score = boxo[8]
    boxr = boxo[0:8].reshape(-1, 2)
    box_area = area( boxr.reshape(8) )

Is the score variable here, in eval.py, refer to a confidence value? Thank you!

[MichalBusta]

Yes.

[MichalBusta]

the score is summation of all confidences merged during NMS, see: https://github.com/MichalBusta/E2E-MLT/blob/61df47f3aa5f3ce54633bb28e3c828d7d59afd68/nms/adaptor.cpp#L24

if you want average value, you have to change code to: p.score / p.nr_polys

(but does not reflect anything - we are learning pixel probability with dice loss and for text detection for fully trained network, values are almost always 0 and 1 )

[shin7]

Thank you!