New Feature - Segmentation labeling


For segmentation labeling, labelbox as it is now presents several challenges.

First, 100% coverage of the image is hard, have to make sure every vector is snapped to a neighboring one.
Second, overlapping polygons lose their order when exported. I tried using “bring to front” and similar options to categorize the entire image with a background classification, then add spot polygons where other classifications existed, but the export puts polygons in class id order, so the ‘bring to front’ polygons can get buried and lost.
Third, a ‘paintbrush’ style tool would be really useful for segmentation labeling


Hi @devin.taylor, thanks for sharing the insights. I believe you can export polygons in vector format and then create masks during post-processing. We actually have paintbrush, superpixels but currently only available in paid tiers. We are soon releasing a comprehensive semantic segmentation tool for everyone in about a month.