Labels are exported in CSV, JSON, COCO, VOC, or Object Detection TFRecord for tensorflow users. To convert your exported data to COCO or Pascal VOC you must first choose the JSON Export Type and the WKT Label Format. Once you've exported your data, follow these Instructions to run the conversion here.
For geometric labels (segmentation, key point, etc...) the coordinates are exported in X-Y coordinates or Well-Known Text (WKT strings).
The coordinate origin for geometric labels is the top left point of the image.
For parsing the WKT format, We recommend using python Shapely.
from shapely.wkt import loads as wkt_loads polygon_wkt= wkt_loads(polygon)
Labelbox can also export image masks from labels created using the image
segmentation tools. Using image masks often means less data preparation is
needed to use labels when training machine learning models.
Labelbox generates an image mask per label class (i.e. if there are 2 car and 2
tree labels in an image, 2 image masks in total - a car and a tree - will be
created). White pixels in the image mask represent the object.