Instance Segmentation Mean Average Precision

[cgrtrifork]

Search before asking

• I have searched the Supervision issues and found no similar feature requests.

Description

Currently the metrics have a MeanAveragePrecision class that works for object detection. It would be great to have an implementation for mAP for masks of an instance segmentation model.

I'd expect only from_detections static method would be used, instead of the from_tensors, without any API change, as the former uses Detections, and these have a mask attribute already.

This goes probably hand in hand with #238.

Use case

No response

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[github-actions]

Hello there, thank you for opening an Issue ! 🙏🏻 The team was notified and they will get back to you asap.

[Bhavay-2001]

Hi @LinasKo, this is the issue I am talking about. Is this available and open to work?

[LinasKo]

Hi @Bhavay-2001,

I've had a look, and we have something more urgent.
How does #1293 sound? (you may reply there)

[kikefdezl]

This is something I thought would work.

I have been debugging for hours trying to figure out why my metrics were producing 0%. It appears that:

from supervision.metrics import MeanAveragePrecision, MetricTarget

metric = MeanAveragePrecision(metric_target=MetricTarget.MASKS)

will always produce metrics that are zero.

I have triple checked and the masks from my targets and predictions overlap almost 1:1.