typed op• Data kinds: lightfield → image
• Call: fullseye.apply(img, "tb_lf_depth_from_focus", a=0.5, b=0.5) (2-D model = एक image + दो scalar knobs a,b∈[0,1])
• GPU: नहीं (केवल CPU reference implementation)
*Figure synthetic 128×128 input पर सच में चलाकर मिला output है। बाईं ओर input, दाईं ओर output। Point clouds ऊपर से देखे गए scatter के रूप में (brightness = z), 1-D series line के रूप में, volumes z दिशा में maximum-intensity projection के रूप में, videos बीच के frame के रूप में, complex images amplitude के रूप में, और जो return values चित्र नहीं बन सकते वे स्वयं values के रूप में दिखाए जाते हैं।*
*Output viridis जैसे pseudo-colour में है (गहरा बैंगनी = कम, पीला = अधिक), ताकि distance, phase, orientation, depth जैसे "मात्राओं के field" पढ़े जा सकें।*
*Knob a output नहीं बदलता (मापा गया: 0.1 / 0.5 / 0.9 पर एक जैसा)।*
*Knob b output नहीं बदलता (मापा गया: 0.1 / 0.5 / 0.9 पर एक जैसा)।*
Stages (पहले आने वाले ops → यह op, बाएँ से क्रम में):
▸ tb_lf_depth_from_focus: stages का figure (docs site)
दूसरी images पर भी (synthetic scene / photo / coins। ऊपर की row में inputs, नीचे की row में उनके outputs। Knobs default पर):
▸ tb_lf_depth_from_focus: दूसरी images पर output (docs site)
refocus sweep में sharpness peak से हर pixel का slope।
> नीचे का विस्तृत विवरण मूल text में है — सारांश और headings का अनुवाद हो चुका है।
Refocus at every slope in *slopes*, measure local sharpness (*measure*:
laplacian = summed modified Laplacian, the classical depth-from-focus operator; variance = local variance; gradient = local gradient energy) in a window x window neighbourhood, and take the slope at which each pixel is sharpest. With subpixel=True (default) the peak is refined by fitting a parabola through the winning sample and its two neighbours on a uniformly spaced sweep — on a non-uniform sweep the refinement is skipped rather than applied with the wrong spacing.
Unbiased where lf_epi_slope is not: measured 2026-09-01 on a 5x5x64x64 synthetic field over a 121-point sweep from -3 to +3, the argmax landed exactly on the true slope in 18 of 18 combinations (true slopes 0.0, +0.5, +1.0, +1.5, +2.0, -1.0 crossed with texture sigma 1.5 / 3.0 / 5.0 px), and the sub-pixel refinement left every one of them unmoved. Its resolution, though, is whatever you put in *slopes* — it cannot see a plane you never refocused on.
That unbiased result holds only for integer and half-integer slopes, and only with interp="cubic" in general. The default "linear" shift blurs a view by an amount that depends on the fractional part of its shift, so the sharpness peak is pulled toward slopes whose per-view shifts are whole pixels — mostly toward 0. Measured 2026-11-04 on Fourier-exact fields (64x64, texture sigma 1.5 / 3.0 px, 61-point sweep -1.5..1.5, median of the map): with 7x7 views a true slope of +0.17 came back as +0.019 (bias -0.151) and -0.16 as -0.018, with 15x15 views +0.05 came back as +0.011; interp="cubic" kept every bias within 0.0037 over 28 cases, at 3.75x the time (43.4 s vs 11.6 s in total). The default stays "linear" because on noisy integer-slope fields cubic is not better (examples/poc_compound_eye.py: RMSE 0.010 linear vs 0.029 cubic px/view) and other shipped examples rely on it. Pass interp="cubic" whenever the slopes you measure are fractional — that is, for every real camera. tests/test_lightfield.py pins both numbers so this note cannot go stale.
Returns (slope_map, sharpness): the (H, W) map of estimated slopes (in px per angular step) and the (H, W) peak focus-measure value, which is the honest confidence — a textureless pixel has no sharpness peak, gets an essentially arbitrary slope, and its sharpness is ~0. Threshold on it rather than trusting the map everywhere.
Raises ValueError: *lf* not a valid light field, *slopes* empty /
over MAX_STACK_SLICES / over MAX_STACK_ELEMENTS / containing a non-finite or over-large value, an even or non-positive *window*, unknown *measure* / *interp* / *edge*.
Typed bridge of the lightfield op lf_depth_from_focus into the 2-D evolution registry: the same implementation, called under the op(v, a, b) convention. This op has no tunable parameter; a and b are unused.
• Sample data catalog (DL URL / license) — 2-D के लिए skimage.data (BSD/public) + synthetic, 3-D के लिए real data sources (Stanford/PDS आदि) के DL URL।
• Operators का स्रोत और संदर्भ — इस op family के मूल research/methods के स्रोत।
नीचे का program सच में चलता है, यह जाँचा गया है (figure वाला ही input)। Studio की help में यह block buttons बन जाता है, जिनसे इसे वहीं load करके चलाया जा सकता है।
img_to_lightfield 0.50 0.50 tb_lf_depth_from_focus 0.50 0.50
▸ यह pipeline load करें · Load करके चलाएँ
नीचे के उदाहरण मूल ledger op lf_depth_from_focus को call करते हैं। यह bridge op वही implementation है, बस fn(v, a, b) convention के अनुसार ढाला गया है, इसलिए व्यवहार वैसा ही लागू होता है (केवल call का रूप अलग है)।
• lightfield_depth — py -3.11 examples/lightfield_depth.py
• plenoptic_decode_pipeline — py -3.11 examples/plenoptic_decode_pipeline.py
• poc_compound_eye — py -3.11 examples/poc_compound_eye.py
• poc_lightfield_depth — py -3.11 examples/poc_lightfield_depth.py
• poc_plenoptic_camera — py -3.11 examples/poc_plenoptic_camera.py
image को input के रूप में लेते हैं)identity · gaussian · mean_box · mean_box_wh · bilateral · kuwahara · unsharp · median
typed)tb_points_to_voxel · tb_estimate_point_normals · tb_iss_keypoints · tb_project_points · tb_render_point_depth · tb_points_render · tb_statistical_outlier_removal · tb_radius_outlier_removal
*Provenance: ops.py — 2D operator registry. यह per-op note tools/opdocs.py md से अपने-आप generate होता है (हाथ से edit न करें)।*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.