Evaluate segmentation quality — sg_evaluate

Computes standard instance segmentation metrics by comparing a predicted mask against a ground truth mask.

Usage

sg_evaluate_segmentation(
  predicted,
  ground_truth,
  metrics = c("dice", "jaccard", "aji", "panoptic_quality", "ap50", "ap75",
    "f1_detection")
)

Arguments

predicted: An sg_mask object with predicted segmentation.
ground_truth: An sg_mask object with ground truth segmentation.
metrics: Character vector of metrics to compute. One or more of "dice", "jaccard", "aji", "panoptic_quality", "ap50", "ap75", "f1_detection". Default is all.

Value

An sg_eval S3 object (a tibble with columns metric and value) with a custom print method.

Examples

pred_labels <- matrix(0L, nrow = 20, ncol = 20)
pred_labels[3:8, 3:8] <- 1L
pred_labels[12:18, 12:18] <- 2L
gt_labels <- matrix(0L, nrow = 20, ncol = 20)
gt_labels[3:9, 3:9] <- 1L
gt_labels[12:17, 12:17] <- 2L
pred <- new_sg_mask(pred_labels)
gt <- new_sg_mask(gt_labels)
result <- sg_evaluate_segmentation(pred, gt)
#> ✔ Computed 7 segmentation metrics.
print(result)
#> <sg_eval> — Segmentation Evaluation
#> ────────────────────────────────────────────────────────────────────────────────
#> dice: 0.8471
#> jaccard: 0.7347
#> aji: 0.7347
#> panoptic_quality: 0.7347
#> ap50: 1
#> ap75: 0
#> f1_detection: 1
#> ────────────────────────────────────────────────────────────────────────────────