Mask_rcnn: How to calculate metrics such a accuracy speficicity and sensitivity

[AmanitaCreaciones]

Hi,

I've been experimenting with Mask RCNN and I would like to know with other metrics you use to see how good it performs more than just loss.

when classificating with bounding boxes I could somehow calculate speficicity and sensitivity depending on the distance between the predicted bounding box and the real object to locate, but with masks I'm not sure how to do it.

Any advice?

[fastlater]

You check the accuracy (mAP) of your model in the inspect model notebook.
The lower the loss, the higher the accuracy. For me, that is enough to know how good my trained model is during training.

[eyildiz-ugoe]

That's actually not enough, you may need to look into COCO metircs. But the authors did not provide any code to evaluate on that, despite of the fact they presented the results in the paper using those metrics.