K Vecs → {centroid, rotation (angle of point[0] about centroid), scale
(its distance from centroid)}. Each write is a closed-form transform about
the centroid (translate / rotate / scale), so the three are cross-channel
invariant. A collapsed cluster makes rotation singular and scale a no-op.
K Vecs → {centroid, rotation (angle of point[0] about centroid), scale (its distance from centroid)}. Each write is a closed-form transform about the centroid (translate / rotate / scale), so the three are cross-channel invariant. A collapsed cluster makes rotation singular and scale a no-op.