Skip to content

annotation_quality_check

Inspect YOLO labels for common annotation mistakes before a bad dataset wastes training time.

Inputs

  • dataset_path
  • optional checks list

Outputs

  • total image count
  • aggregate issue counts by rule
  • representative issue rows with file paths and messages

Checks

  • invalid class ids
  • empty label files
  • tiny bounding boxes
  • out-of-bounds YOLO coordinates
  • near-identical overlapping boxes
  • extremely crowded images

Examples

{ "dataset_path": "~/data/warehouse" }

Common Errors

  • dataset path missing
  • non-YOLO datasets should be converted before running this tool

dataset_inspect, dataset_validate, eval_error_analysis