CircleNet in Pytorch?

I’m working on object detection in a domain where bounding cirlces would make more sense than boxes, and noticed that CircleNet appears to do this. Has anyone in here tried this? If so, does it work well, and is it stable?