Skip to content

eval_run

Run a validation pass on a checkpoint.

Inputs

  • checkpoint
  • dataset_path
  • split
  • batch, imgsz, device
  • conf, iou, task

Outputs

  • aggregate metrics such as mAP50, precision, and recall
  • per-class stats
  • evaluation duration

Examples

{"checkpoint":"~/runs/run_demo/weights/best.pt","dataset_path":"~/data/mini_yolo"}
{"checkpoint":"run_demo","dataset_path":"~/data/mini_yolo","split":"val"}
{"checkpoint":"~/models/baseline.pt","dataset_path":"~/data/retail","device":"cpu"}

Common Errors

  • checkpoint missing
  • dataset missing

eval_per_class, eval_error_analysis, eval_compare