Yolo 11 segmentation on CoreML returns all 0 mask prototypes

Support YOLO
,
1

I am exporting the Yolo11n-seg model to Core ML. In this format the model seems to lack any post-processing (e.g. NMS). I can live with that, but…

The output of the Core ML model is 2 tensors: var_1366 and p. The latter is supposed to contain the mask prototypes if I understand correctly, but its contents are always all zeroes. Is this a known issue? Any ideas of what I could be doing wrong? (the mask coefficients and the boxes do seem to contain plausible data)

2

Same issue with yolov8-seg.

3

Are all the values 0?

4

Yes, they are all 0. Seems like a bug in the CoreML export to me.

5

Does the Core ML model work in Ultralytics?

6

What do you mean? CoreML is an Apple framework, which I use from Swift.

7

You can load and run CoreML model in Ultralytics (if you have a Mac)