SED Segmentation

Sagui clusters one photometric SED per supported pixel. The support mask and clustering transform are chosen before measuring the regional fluxes.

Exact Ward

seg <- segment_regions(
  input = cube,
  Ncomp = 20,
  use_starlet_mask = TRUE,
  cluster_pretransform = "none"
)

Exact Ward clustering stores all pairwise distances. Use it only when the supported pixel count fits available memory.

Sparse Ward

seg <- segment_regions_large(
  input = cube,
  Ncomp = 20,
  use_starlet_mask = TRUE,
  cluster_pretransform = "none",
  knn_k = 40,
  auto_k = FALSE
)

The sparse backend restricts candidate merges to a nearest-neighbour graph. Its result depends on knn_k, auto_k, the support mask, and the transform.

Choose transforms

Use cluster_pretransform, not the deprecated pretransform alias. The default "none" preserves the input representation. Alternatives such as "copula_gaussian" change the clustering geometry; compare them against "none".

Region labels are categorical. Their integer values do not represent a physical ordering.