GuidesDatasets and Models

Oriented Bounding Box Inputs

Prepare YOLO four-corner oriented bounding-box annotations and Model output.

An oriented bounding box is a rectangular annotation that may be rotated. Knotest supports the YOLO Oriented Bounding Boxes Dataset format.

Preparation checklist

  • Keep each four-corner annotation in the supported point order.
  • Confirm coordinates correspond to the source image dimensions.
  • Preserve class IDs and label order.
  • Configure the Model for the OBB task rather than axis-aligned detection.
  • Inspect imported previews for orientation and corner-order mistakes.

Use OBB only where space is constrained. Product documentation uses the full term oriented bounding box in explanatory copy.

These tutorial samples demonstrate the oriented-box format, but their visible boxes are nearly axis-aligned. Also test a clearly rotated annotation when validating your own Dataset.

Helmet oriented bounding-box Dataset sample grid with annotation overlays.

Watch: Inspect oriented boxesCompare original samples with their annotation overlays.Silent demo · Subtitles follow page language · Full screen for detail

Segmentation Inputs

Prepare YOLO Segmentation or COCO Instance Segmentation inputs.

Tabular Inputs

Prepare CSV, TSV, or Parquet classification and regression data.

On this page

Preparation checklist