Why roboflow showing precision value more than 100% after running instance segmentation yolo8 model. Which is impracticle.
Hello! I’m an engineer on the Roboflow team. I’m looking into this right now. Are there any other models that are producing this erroneous output for you, or is it just YOLOv8 with instance segmentation?
Following up here in case anyone else might be experiencing the same problem - Roboflow currently supports custom model weights trained using the dependency ultralytics==8.0.196
We are working to resolve inconsistencies in the training results provided by newer versions of ultralytics that result in the incorrect precision metrics displayed in the user interface.