Skip to content

eval_compare

Evaluate multiple checkpoints on the same dataset and rank the results.

Inputs

  • checkpoints
  • dataset_path
  • shared evaluation knobs: split, batch, imgsz, device, conf, iou

Outputs

  • ordered checkpoint comparison rows
  • best checkpoint by mAP50 and mAP50-95

Examples

{"checkpoints":["run_a","run_b"],"dataset_path":"~/data/mini_yolo"}
{"checkpoints":["~/models/a.pt","~/models/b.pt"],"dataset_path":"~/data/retail"}
{"checkpoints":["run_small","run_large"],"dataset_path":"~/data/factory","device":"cpu"}

Common Errors

  • any checkpoint or dataset failure from eval_run

eval_run, run_compare