wiring op• 데이터 종류: labels2d × labels2d × table → table
• 호출: import fullseye as fs; fs.ledger.seg_wiring_variation(a, b, synapses, spacing=None, background=0)(구현을 직접 호출하려면 import segcompare; segcompare.seg_wiring_variation(a, b, synapses, spacing=None, background=0), 원장에서 가져오려면 opsemproof.get("seg_wiring_variation"))
> 이 연산자의 설명은 아직 번역이 없습니다. 원문을 그대로 싣습니다.
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.
• 샘플 데이터 카탈로그(DL URL / 라이선스) —— 2-D 는 skimage.data(BSD/public)+ 합성, 3-D 는 실데이터 소스(Stanford/PDS 등)의 DL URL.
• 연산자의 내력·참고문헌 —— 이 연산자 족의 바탕이 된 연구/기법의 출처.
• 알고리즘의 정전(저자·연도)과 용도는 위의 패밀리 사용 가이드에 적혀 있습니다.
• poc_em_wiring_errors — py -3.11 examples/poc_em_wiring_errors.py
table 를 입력으로 받는 것)seg_synapse_partners · seg_synapse_nri · seg_wiring_exposure
wiring)seg_synapse_partners · seg_synapse_nri · seg_wiring_exposure
*Provenance: segcompare.py — EMPROOF 연산자 레지스트리. 이 op 노트는 tools/opdocs.py md 가 자동 생성합니다(직접 편집하지 마세요).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.