search

aim-uofa / AdelaiDet

AdelaiDet is an open source toolbox for multiple instance-level detection and recognition tasks.
https://git.io/AdelaiDet
Other
3.38k stars 648 forks source link

Whether the recognition metrics have been calculated? #463

Closed Single430 closed 3 years ago

Single430 commented 3 years ago

First of all, thank you very much for your work! I use abcnet to do finetune on the training total text dataset, e2e-hmean, det-hmean basically meets expectations, but I don't know what the metrics are recognition, how much can I provide for the next Acc and norm_ED indicators? @Yuliang-Liu

Yuliang-Liu commented 3 years ago

@Single430 The metrics are following the official implementation of ICDAR 2015, which use full match for a correct recognition result. You may add additional indicators in the evaluation scripts.

Single430 commented 3 years ago

@Single430 The metrics are following the official implementation of ICDAR 2015, which use full match for a correct recognition result. You may add additional indicators in the evaluation scripts.

Thank you very much for replying! But I should make it very clear that I need to recognition metrics, that is, RECOGNITION_ONLY_RESULTS, specifically the accuracy of the editing distance. My side has been achieved, but I don't know if it's right or wrong, so hopefully I can see the official comparison.

Mainly calculated recognition metrics: norm_ED 0.94 Acc 0.5 is very low