fullseye

seg_rand — EMPROOF score op

• データ種: labels2d × labels2d → table
• 呼び出し: import fullseye as fs; fs.ledger.seg_rand(a, b, ignore_label=None) (実装を直接呼ぶなら import segcompare; segcompare.seg_rand(a, b, ignore_label=None)、台帳から引くなら opsemproof.get("seg_rand"))

使い方

Rand index, adjusted Rand index and the adapted Rand error (CREMI / SNEMI).

With n_ij the contingency table, a_i / b_j its row and column sums and N the pixel count:

• rand_index = fraction of pixel pairs on which the two agree (same/different);
• adjusted_rand_index = Hubert & Arabie’s chance-corrected version (1 = identical, about 0 = independent);
• adapted_rand_error = 1 - F over pairs of distinct pixels: pair precision = (sum n_ij^2 - N) / (sum b_j^2 - N) — of the pairs the candidate puts together, the fraction the truth also puts together — and recall = (sum n_ij^2 - N) / (sum a_i^2 - N) (Arganda-Carreras et al. 2015). The - N removes each pixel’s pair with itself. scikit-image’s adapted_rand_error returns the same error, but its precision is divided by the truth pairs (its code sums the rows of the truth-by-test table), i.e. it is this recall; its docstring says otherwise.

Raises ValueError: as :func:seg_contingency; fewer than 2 counted pixels.

詳しい使い方ガイド

• emproof ファミリ ガイド

参考(サンプルデータ・文献)

• サンプルデータ カタログ(DL URL / ライセンス) — 2-D は skimage.data(BSD/public)+ 合成、3-D は実データ源(Stanford/PDS 等)の DL URL。
• 演算子の来歴・参考文献 — この op 族の元になった研究/手法の出典。
• アルゴリズムの正典(著者・年)と用途は上記ファミリ使い方ガイドに記載。

実行できる例(この op を実際に呼ぶ検証済みサンプル)

• poc_em_split_merge_score — py -3.11 examples/poc_em_split_merge_score.py

型が繋がる次の op(table を入力に取れる)

seg_synapse_partners · seg_wiring_variation · seg_synapse_nri · seg_wiring_exposure

同カテゴリ(score)

seg_contingency · seg_variation_of_information

---

Provenance: segcompare.py — EMPROOF operator registry. この per-op ノートは tools/opdocs.py md が自動生成(手編集しない)。

© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.

This site is open source. Improve this page.