search

facebookresearch / detectron2

Detectron2 is a platform for object detection, segmentation and other visual recognition tasks.
https://detectron2.readthedocs.io/en/latest/
Apache License 2.0
30.46k stars 7.48k forks source link

Implementation of mAP and F1 score at different IoU #3733

Open Nurungyi opened 2 years ago

Nurungyi commented 2 years ago

🚀 Feature

A clear and concise description of the feature proposal.

Motivation & Examples

Now, in Detectron2, AP of each class is calculated by IoU @ 0.50:0.95. Although i tried to find the relevant comment from this github, I could only find this (https://github.com/facebookresearch/detectron2/issues/1877) But, maybe most of people prefer the implementation of Detectron2 on googlb colab or something like that. It may be difficult to change it on Colab and quickly evaluate it. If there is easier method for mAP calculation of different IoUs on Google colab, please let me know.

And also, for many real applications, except for mAP, F1 score is still widely used and many metrics including mAP and F1 score are evaluated at different IoU threhold, which is also one parameters for a practical use of Faster R-CNN through Detectron2.

So, I think that implementation of mAP and F1 score at different IoU will be practically and prevalently used.

aymanaboghonim commented 2 years ago

Have you learnt how to implement F1 score ??