I’m using a workflow hosted on roboflow for my YoloV 11 instance segmentation custom model.
My workflow:
I’ve noticed in the payload for each prediction there is only a single list of points.
When I test the workflow I have an image where one of the classes is obscured by another class. (green “Tunnel 6m” at the bottom left) When using yolov11 api directly it returns two masks for this prediction and sv.MaskAnnotator correctly outlines both parts of the mask.
However, in the payload from the hosted inference workflow only the points for the larger mask are returned.
In the following image you can see the green bounding box and centroid for the “Tunnel 6m” class correctly outline the entire curved tunnel. I’m assuming that is because the bounding box/centroid are computed using all the masks returned by the model for each prediction.
But the polygon visualizer only outlines the larger mask’s points because the API only returns a single list of points for each prediction(?)
Is this a known issue with the Roboflow workflow API for instance segmentation results or am I misconfigured in some way where I don’t receive either an additional prediction or a list of points for each mask?
Thanks,
Steve