eval_error_analysis
Inspect confusion patterns and worst examples beyond headline metrics.
Inputs
checkpointdataset_pathsplittop_nimgsz,device,conf,iou
Outputs
- confusion entries
- top error samples populated from saved validation predictions and ground-truth labels
- false-positive and false-negative counts
- evaluation duration
Examples
{"checkpoint":"run_demo","dataset_path":"~/data/mini_yolo"}
{"checkpoint":"~/models/retail.pt","dataset_path":"~/data/retail","top_n":20}
{"checkpoint":"~/models/factory.pt","dataset_path":"~/data/factory","conf":0.15}
Common Errors
- missing dataset or checkpoint
- malformed or unsupported validation output from the underlying Ultralytics run
Related Tools
eval_run, eval_compare, run_compare