Create an annotation task for human review — sg_create_annotation

Samples informative image patches for manual annotation or correction. Patches are selected on a grid or, when an initial mask is provided, prioritised by segmentation uncertainty.

Usage

sg_create_annotation_task(
  image,
  initial_mask = NULL,
  region = NULL,
  n_patches = 20L,
  patch_size = 256L
)

Arguments

image: An sg_image object.
initial_mask: Optional sg_mask object. When provided, patches are scored by segmentation uncertainty and sampled accordingly.
region: Optional numeric vector c(row_min, row_max, col_min, col_max) to restrict patch sampling to a sub-region.
n_patches: Integer. Number of patches to sample. Default 20L.
patch_size: Integer. Size of each square patch in pixels. Default 256L.

Value

An sg_annotation_task S3 object (a list with elements patches, image_id, n_patches, patch_size, and created).

Examples

pixels <- matrix(runif(100 * 100), nrow = 100, ncol = 100)
img <- new_sg_image(pixels)
task <- sg_create_annotation_task(img, n_patches = 4L, patch_size = 32L)
#> ✔ Created annotation task with 4 patches.
#> ℹ Patch size: 32 x 32 px.
print(task)
#> <sg_annotation_task>
#> Patches: 4 (32 x 32 px)
#> Created: 2026-08-22 13:08:29