Project Type:
Instance Segmentation (YOLOv11)
Operating System & Browser:
(Windows 11 / Google Chrome)
Project Universe Link or Workspace/Project ID:
(Please fill in: link to project or workspace/project ID)
Hello everyone,
I am facing a serious problem regarding training of YOLOv11-instance-segmentation model on roboflow. My test set contains just two images to simulate this error.
After training of YOLOv11-instance-segmentation model finished it shows following metrics on test set:
mAP@50 - 97.4%
Precision - 98.2%
Recall - 95.9%
However, when visualizing the prediction on the test set image, it clearly failed to segment one piece of cutlery entirely, and the masks are not very accurate around the edges. Based on this, the test metrics cannot be so high. Same issue is with another image from test set. Here is comparison of annotated image from test set and output of trained model YOLOv11. Even in the bottom part you can see it shows 100% mAP50 on cutlery which is obviously wrong:
I would really appreciate any explanation. Thanks in advance!