I would like to use the Ground Truth Labeler App to create labels for semantic segmentation. That means I need arbitrary shaped regions, so my label type is 'Pixel Label'.
Unfortunately, the existing PointTracker algorithm only tracks rectangular bounding boxes. I have developed my own script which will track an arbitrary shaped polygon, and have formatted it similarly to the driving.vision.labeler.PointTracker class. But, when I import this algorithm to the Ground Truth Labeler App, the pixel labels are still greyed out, and I can only select rectangular labels to track. I have modified this function:
function isValid = checkLabelDefinition(~, labelDef)
isValid = labelDef.Type==labelType.PixelLabel;
What else do I need to do?
While I am testing my new function, I am not sure if modifications are being picked up by the Ground Truth Labeler App. For example, I have modified the description in the comments at the top of the class, but the text in the drop-down menu stays the same, even after refreshing. Is there a way to remove an algorithm from the list, so that I can be sure it is really refreshing?