How to fine‑tune YOLOv8 with Roboflow turfgrass disease images (and which model size)?

[acayci]

Hi Roboflow Community,

I’ve downloaded a turfgrass disease dataset from Roboflow Universe and want to fine‑tune YOLOv8 for object detection.

Questions:

How would I start? Is there a step by step process documented for the datasets downloaded from Roboflow Universe?
Export: best format from Roboflow (YOLOv8 vs COCO) and recommended augments?
Splits: tips for small/imbalanced classes?
Training cmd/hparams (img, epochs, batch, lr)?
Evaluation: mAP50/95, PR curves, test/inference.
Model size (YOLOv8x?): Is v8x appropriate here?

Thank you for your help,
Mustafa