Quickshift¶
Quickshift is a mode-seeking algorithm: it estimates a density over the same combined position+color feature space SLIC uses, then moves each pixel uphill toward its nearest neighbor with higher density, tracing out a tree of parent pointers. Cutting that tree at a distance threshold yields the final segments. Because segments are built by following density gradients rather than a fixed grid or a single global threshold, Quickshift naturally produces a hierarchy of consistent, perceptually uniform segments — you can obtain coarser or finer segmentations from the same computed tree just by changing the cut distance.
It’s more computationally expensive than SLIC but doesn’t require choosing the number of output regions in advance.
Accessible via spice.clustering.quickshift_clustering().
See also
SLIC Superpixels — a faster superpixel algorithm over the same feature space
Felzenszwalb — another approach that doesn’t need a region count in advance
Image Clustering — every other clustering algorithm