wiring op• 数据种类:labels2d × table → table
• 调用: import fullseye as fs; fs.ledger.seg_wiring_exposure(labels, synapses, spacing=None, background=0, p=0.5)(要直接调用实现,import segcompare; segcompare.seg_wiring_exposure(labels, synapses, spacing=None, background=0, p=0.5);从台账取用则 opsemproof.get("seg_wiring_exposure"))
> 该算子的说明尚无译文,以下照原文给出。
How much wiring each object stakes on staying one piece — from synapse end counts alone.
Cutting an object whose `s synapse ends (synapses as in :func:seg_synapse_partners`)
fall `s1 / s - s1 on the two sides costs exactly (s / 2n) * H2(s1 / s)` bits of
ends-level split (:func:seg_wiring_variation), and a hidden merge of two neurons inside
one object costs the same expression as merge. So what an object can cost is fixed by its
end count before any geometry is known:
• `bound = s / 2n` — the supremum over all cuts (reached when the ends halve). The bounds
sum to `n_ends / 2n: exactly 1 bit when no end lies on background`.
• `worst = (s / 2n) * H2(floor(s / 2) / s)` — the largest cost a cut can actually reach.
• `expected = (s / 2n) * E[H2(K / s)], K ~ Binomial(s, p)` — the cost of a cut that
puts each end on the cut-off side independently with probability `p` (default 1/2,
the cut through the object's median voxel with ends spread evenly over it). This is the
count-only prediction; the gap to the observed cost is the spatial clustering of the ends.
• `density = (s / 2n) / (voxels / N)` — the object's share of ends over its share of the
`N` voxels of the volume. For an edit, wiring bits / pixel bits = density × (H2 of the
end fractions / H2 of the voxel fractions): merges have the second factor near 1 (ends
follow voxels), so their ratio is the density; a cut's second factor is
`H2(s1 / s) / 1` and is what the binomial predicts.
Objects with no end are not listed. Returns `labels, voxels, ends, density`,
`bound, worst, expected (aligned arrays), n synapses, n_ends` kept,
`n_background ends dropped, total_bound, total_expected`.
Raises `ValueError: as :func:seg_synapse_partners; p` not strictly inside
(0, 1); no end outside the background.
• 示例数据目录(下载 URL / 许可证) —— 2-D 用 skimage.data(BSD/公有领域)加合成图,3-D 给出真实数据源(Stanford/PDS 等)的下载 URL。
• 算子来历与参考文献 —— 该算子族所依据的研究/方法出处。
• 算法的正典(作者・年份)与用途见上面的族使用指南。
• poc_em_wiring_errors — py -3.11 examples/poc_em_wiring_errors.py
table 作为输入)seg_synapse_partners · seg_wiring_variation · seg_synapse_nri
wiring)seg_synapse_partners · seg_wiring_variation · seg_synapse_nri
*Provenance: segcompare.py — EMPROOF 算子登记表。本条目由 tools/opdocs.py md 自动生成(请勿手工编辑)。*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.