Model type: Roboflow 3.0 Instance Segmentation (Accurate)
Vision: Instance segmentation
I need help interpreting the difference between these loss graphs. Both reached the same final result, but I don’t understand why one of them looks much smoother (nicer), while the other one looks rough.
In the first graph (nicer), I used transfer learning from the MSCOCO dataset provided by Roboflow. In the second graph (the rough one), I used transfer learning from a checkpoint of a previously generated model with an mAP of 89%.