fullseye

patchbank_fit — SPC anomaly_patch op

• データ種: video → table
• 呼び出し: import fullseye as fs; fs.ledger.patchbank_fit(stack, patch=8, stride=4, scales=(1.0, 2.0, 4.0), groups=('gradient', 'gabor', 'lbp'), n_orient=4, coreset=0.1, proj_dim=32, position_weight=0.0, seed=0) (実装を直接呼ぶなら import anomaly_patch; anomaly_patch.patchbank_fit(stack, patch=8, stride=4, scales=(1.0, 2.0, 4.0), groups=('gradient', 'gabor', 'lbp'), n_orient=4, coreset=0.1, proj_dim=32, position_weight=0.0, seed=0)、台帳から引くなら opsspc.get("patchbank_fit"))

使い方

Build a memory bank of good patch features, thinned by greedy k-centre covering (table).

stack is (n, H, W) or (n, H, W, 3) of good images (n >= 1). Features are standardised with the mean and the standard deviation of each dimension over all good patches. With position_weight > 0 the grid position (0..1 in each direction) times position_weight * sqrt(D) is appended, so a patch is compared with good patches from nearby positions (0 = position-free). coreset in (0, 1] is the fraction of patches kept: the greedy k-centre cover picks, from a seeded start, always the patch farthest from the ones already kept, measuring distances in a seeded random projection to proj_dim dimensions (None = exact). The covering radius in that space after the selection is returned: the largest distance from any good patch to the bank. coreset=1 keeps everything (radius 0, no selection, so seed and proj_dim have no effect).

Returns the model: bank (m, D’), mean / std (D), n_total patches, n_memory, fraction, covering_radius, the feature parameters and image_shape.

Raises ValueError for invalid parameters, a feature that is constant over all good patches, or a stack that would exceed the memory guard.

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_patch_anomaly — py -3.11 examples/poc_patch_anomaly.py

型が繋がる次の op(table を入力に取れる)

msa_anova_table · msa_gauge_rr · msa_bias_linearity · msa_attribute_agreement · gum_standard_uncertainty · gum_propagate · gum_expanded · gum_monte_carlo

同カテゴリ(anomaly_patch)

patch_features · patchdist_fit · patchdist_score · patchbank_score

---

Provenance: anomaly_patch.py — SPC 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.