Filtering the detection about autodistill HOT 6 CLOSED

Hello @Rahul-Pi! Thank you for this note. The supervision.non_max_suppression function (https://roboflow.github.io/supervision/detection/utils/#non_max_suppression) may be helpful in filtering out duplicate detections.

@yeldarby Should we do NMS by default on object detection results?

from autodistill.

@capjamesg you can also call .with_nms on detections.

from autodistill.

Thank you @capjamesg and @SkalskiP

Using base_model.predict(test_image).with_nms(threshold=0.85, class_agnostic=True) solved the problem.

Would it be possible to have this as an argument in the DetectionBaseModel.label?

Should I create a PR with this change?

from autodistill.

Using base_model.predict(test_image).with_nms(threshold=0.85, class_agnostic=True) solved the problem.

So awesome to hear that! 🔥

from autodistill.

@yeldarby Should we do NMS by default on object detection results?

Yes, I think we should.

from autodistill.

My apolgoies for necroing this but is it possible to apply this to GroundedSAM through AutoDistill? I dont want to feed Yolo bad segementation masks and ideally less tedious work is better

from autodistill.