DataSet method "segmentation_labels" from dataset.py returning empty list.

The method [segmentation_labels] from the class DataSet should return all the unique classes labels contained in the segmentation masks. Instead, it return an empty list which is useless considering that it is misleading. It should be defined according to the segmentations folder linked to the dataset object (check if there is segmentations and check for unique values contained inside of each image) and add each unique class value inside a set.

def segmentation_labels(self) -> List[Any]:
        return []

It should at least raise an error if there is no segmentations linked to the object.

Start in opensfm/dataset.py at DataSet.segmentation_labels and inspect how the dataset's segmentations folder is located. Check the available segmentation masks for unique class values and determine the expected behavior when no segmentations are linked. Done means the method returns the collected unique labels instead of an empty list and handles the missing-segmentation case explicitly.