fullseye

seg_wiring_variation — EMPROOF wiring op

• データ種: labels2d × labels2d × table → table
• 呼び出し: import fullseye as fs; fs.ledger.seg_wiring_variation(a, b, synapses, spacing=None, background=0) (実装を直接呼ぶなら import segcompare; segcompare.seg_wiring_variation(a, b, synapses, spacing=None, background=0)、台帳から引くなら opsemproof.get("seg_wiring_variation"))

使い方

How a candidate segmentation b corrupts the wiring diagram of truth a, in bits.

Two levels, both VOI split / merge (synapses as in :func:seg_synapse_partners):

• synapse ends (split / merge, voi): the segmentation VOI measured only at the 2n synapse ends instead of every pixel. split = two ends on one true neuron now on different objects (the neuron was cut between its synapses); merge = ends on different true neurons now on one object (they were glued). Cutting a neuron whose s ends divide s1 / s2 raises split by exactly (s / 2n)·H2(s1 / s). Pixels far from any synapse cost nothing here.
• connections (connection_split / connection_merge): synapses grouped by their (pre object, post object) pair — one true connection scattered over several, or different true connections fused into one (synapses attributed to the wrong partner). A connection with a single synapse cannot scatter, so this level is blind to cuts that only rename a one-synapse connection; the ends level is not.

Provenance: the ends level is the “synapse VI” of Plaza, Scheffer & Chklovskii 2014 (Focused proofreading, arXiv:1409.1199; VI over the voxels carrying synaptic annotations, implemented in NeuroProof, C++). The connection level, the lost-synapse convention and the per-edit closed forms ((s/2n)·H2(s1/s) for a cut) are this module’s. Synapses with an end on background in the truth are dropped; in the candidate, each such synapse is lost and counted in n_lost as its own singleton at both levels (not one shared “background” object, which would fake a huge merge). Also returns split_connections (true connections spread over 2+ candidate ones), merged_connections (candidate connections holding synapses of 2+ true ones), edges_a / edges_b and n (synapses scored).

Raises ValueError: as :func:seg_synapse_partners; shapes of a and b differ; no synapse left after dropping background in the truth.

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_wiring_errors — py -3.11 examples/poc_em_wiring_errors.py

型が繋がる次の op(table を入力に取れる)

seg_synapse_partners · seg_synapse_nri · seg_wiring_exposure

同カテゴリ(wiring)

seg_synapse_partners · seg_synapse_nri · seg_wiring_exposure

---

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.