MichalBusta / E2E-MLT

E2E-MLT - an Unconstrained End-to-End Method for Multi-Language Scene Text
MIT License
291 stars 84 forks source link

How to evaluate E2E-MLT? #63

Closed MiTeng0215 closed 4 years ago

MiTeng0215 commented 4 years ago

In the binary classification task, I know the meaning of 'precision' and 'recall', but how do I calculate 'precision' and 'recall' in the E2E evaluation criteria? How to represent TP and FP?Such as the End-to-end evaluation on generic setting of ICDAR 2015 dataset.

MichalBusta commented 4 years ago

Hi, it is same like in binary classification: in ICDAR competition protocol, you have to pass detection threshold 0.5 (IoU) and if your transcription matches GT you get the cookie (TP). Everything else is FP. (with one exception- there are also "do not care regions")

for details see https://rrc.cvc.uab.es/?ch=4&com=tasks and referenced papers.