depth op• データ種: なし → table(引数だけで決まる op —— 画像やデータの入力を取らない)
• 呼び出し: import lightfield; lightfield.lf_plenoptic_design(focal_mm=50.0, f_number=8.0, object_mm=300.0, pixel_um=3.45, mla_pitch_um=27.6, sensor_px=(2048, 2448), *, subpixel_px=0.1) (または opslightfield.get("lf_plenoptic_design"))
プレノプティックカメラの寸法決め: 角度/空間分解能と深度範囲がいくら買えるか。
> 以下の詳細説明は原文のままです —— 要約と見出しは訳出済み。
The plenoptic trade in one table. A microlens spanning ``mla_pitch_um /
pixel_um`` pixels turns that many pixels into that many *directions*, so the
sensor's pixel count is unchanged but the image is `U*V` times smaller and
carries `U*V viewpoints. This operator composes :mod:optics` rather than
re-deriving it: `optics.thin_lens` places the image, and
`optics.depth_of_field` is called twice — once with the pixel pitch as
the circle of confusion (the depth of field of a single refocused slice) and
once with the *microlens* pitch (the range over which refocusing can still
recover a sharp image). Their ratio is the refocusing gain, and it comes out
at the angular resolution, which is the textbook result — measured
2026-09-01 at `f = 50 mm, N = 8, s_o = 300 mm`: an 8x8 angular
grid gives `refocus_gain = 8.0038, 10x10 gives 10.0075` and 6x6 gives
`6.0016`.
Returns a dict — `angular_u / angular_v` (whole pixels per microlens,
from `floor) · angular_exact` (the unrounded ratio) and
`pitch_is_integer` (whether the MLA pitch is a whole number of pixels; it
usually is not, which is why real decoding needs sub-pixel calibration) ·
`spatial_w / spatial_h` (microlenses = sub-aperture image size) ·
`n_views · resolution_loss (U*V) · image_mm` /
`magnification / working_distance_mm` from the thin lens ·
`aperture_mm (f/N) · baseline_mm` (viewpoint spacing across the
pupil, `aperture / (U - 1)) · focal_px_subaperture` (focal length in
units of the *microlens* pitch, the pixel of a sub-aperture image) ·
`dof_pixel_mm / dof_refocus_mm and refocus_gain` · and
`depth_precision_mm`, the object-side distance change that moves the
disparity by *subpixel_px* pixels (`Z^2 * dp / (focal_px * baseline)`) —
the honest depth resolution at *object_mm*.
Raises `ValueError`: any non-positive or non-finite length, an
`mla_pitch_um` smaller than two *pixel_um* (fewer than 2 directions is not
a light field), a sensor smaller than one microlens, a non-positive
*subpixel_px*, `object_mm == focal_mm` (propagated from
`optics.thin_lens`: the object images at infinity), and an angular
resolution of 1 in either axis, where the baseline would be a 0/0.
• 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。
• lightfield_depth — py -3.11 examples/lightfield_depth.py
table を入力に取れる)—
depth)lf_depth_from_focus · lf_epi_slope · lf_disparity_to_depth · lf_all_in_focus
*Provenance: lightfield.py — LIGHTFIELD 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.