Export

Exports a generated dataset to standard formats: COCO, YOLO and Pascal VOC.

Run it once after generating, on the generating path. It reads the <index>.json labels; only datapoints with an image (a Render step) and with bboxes or instance masks are exported. It works without Blender. COCO with masks needs numpy and OpenImageIO, which come with Blender.

coco

coco(
    path: str,
    output: Optional[str] = None,
    classes: Optional[Sequence[str]] = None,
    bbox_from: Literal["label", "mask"] = "label",
) -> str

Writes a COCO detection and instance segmentation JSON.

Masks are stored as uncompressed RLE from the instance masks. Instances whose mask is empty (fully hidden), or was not written because it was empty (Segmentation(skip_empty=True)), are skipped. Each annotation also has an "objects" field with the object names.

Parameters:

• path (str) –

  generating path with the <index>.json labels. // paths are relative to the .blend file when running in Blender.

• output (Optional[str], default: None ) –

  JSON file to write. None = <path>/coco.json.

• classes (Optional[Sequence[str]], default: None ) –

  class names in category id order (ids start at 1). None = in order of first appearance.

• bbox_from (Literal['label', 'mask'], default: 'label' ) –

  "label" for the boxes of the BBox step, which include hidden parts, or "mask" for the extent of the visible pixels in the instance masks.

Returns:

• str –

  Path of the written file.

Raises:

• ValueError –

  an unknown bbox_from, or a class in the labels missing from classes.

Example

export.coco("//dataset", classes=["car", "table"])

yolo

yolo(
    path: str, classes: Optional[Sequence[str]] = None
) -> str

Writes YOLO detection labels.

Writes <index>.txt next to every image, with one class x_center y_center width height line per bbox (normalized to 0-1), classes.txt, and dataset.yaml pointing at the folder, ready for Ultralytics.

Parameters:

• path (str) –

  generating path with the <index>.json labels.

• classes (Optional[Sequence[str]], default: None ) –

  class names in class id order (ids start at 0). None = in order of first appearance.

Returns:

• str –

  Path of dataset.yaml.

Raises:

• ValueError –

  a class in the labels missing from classes.

Example

export.yolo("//dataset")

voc

voc(path: str, output_dir: Optional[str] = None) -> str

Writes Pascal VOC XML annotations, one file per image, with 1-based pixel bboxes.

Boxes touching the image border are marked truncated.

Parameters:

• path (str) –

  generating path with the <index>.json labels.

• output_dir (Optional[str], default: None ) –

  folder for the XML files. None = <path>/Annotations.

Returns:

• str –

  The output folder.

Example

export.voc("//dataset")