Skip to contents

Create sparse semantic segmentation masks from point labels

make_sparse_masks.Rd

Writes sparse masks where point neighborhoods contain class_id values and all unlabeled pixels are ignore_index. These are weak labels, not dense human-annotated segmentation masks.

Usage

make_sparse_masks(
  points,
  image_root,
  out_dir,
  radius = 3,
  ignore_index = 255,
  background_index = 0,
  class_col = "ml_class"
)

Arguments

points

A tidy point table.

image_root

Image root used to match images when needed.

out_dir

Output directory.

radius

Disk radius in pixels around each point.

ignore_index

Pixel value for unlabeled pixels.

background_index

Reserved background value. Included for downstream schemas; unlabeled pixels still default to ignore_index.

class_col

Label column used to assign class_id values when they are missing. Defaults to ml_class, with automatic fallback to raw CPCe labels for bare workflows without a crosswalk.

Value

A mask manifest tibble.

Examples

example_dir <- system.file("extdata", package = "pointcoral")
pts <- read_cpce_file(file.path(example_dir, "HIW_158_W_U-1.cpc"))
out_dir <- file.path(tempdir(), "pointcoral-masks-example")
make_sparse_masks(
  pts[1:3, ],
  image_root = example_dir,
  out_dir = out_dir,
  radius = 2,
  class_col = "raw_label"
)
#> # A tibble: 1 × 9
#>   image_path         image_id mask_path image_width image_height n_points radius
#>   <chr>              <chr>    <chr>           <int>        <int>    <int>  <dbl>
#> 1 /home/runner/work… HIW_158… /tmp/Rtm…         900          566        3      2
#> # ℹ 2 more variables: ignore_index <int>, background_index <int>