polarization op• 데이터 종류: polsweep → image2d
• 호출: import fullseye as fs; fs.ledger.polarization_dolp_map(images, angles_deg=(0.0, 45.0, 90.0, 135.0), max_violation_frac=0.0)(구현을 직접 호출하려면 import specularity; specularity.polarization_dolp_map(images, angles_deg=(0.0, 45.0, 90.0, 135.0), max_violation_frac=0.0), 원장에서 가져오려면 opsspecular.get("polarization_dolp_map"))
• GPU: 없음(CPU 참조 구현만)
화소별 직선 편광도. → [0, 1] 의 (H, W).
> 아래 상세 설명은 원문입니다 —— 요약과 제목은 번역되어 있습니다.
sqrt(S1^2 + S2^2) / S0, from the same sinusoid fit as polarization_separate. It answers the question that decides whether a polariser is worth mounting at all: 1 means the light at that pixel is fully linearly polarised and a crossed analyser extinguishes it completely, 0 means the polariser will only cost you a stop of exposure.
Being a ratio, it is invariant to exposure: scaling every frame by 1e-4, 1e4 or 1e8 moves it by at most 3.9e-16 (measured), which is what makes it comparable across a production line where the lamps age. Not *bit*- identical — the least-squares fit rounds differently at a different scale — and the difference between "invariant" and "bit-identical" is the kind of claim this library measures rather than asserts.
Pixels with zero total radiance have no degree of polarisation (every ratio would be 0/0) and are returned as 0.0 rather than as NaN; that is a convention, and it is stated here because a NaN spreading into a thresholding routine is exactly the silent failure this library refuses.
★Dark pixels (2026-11-09): "zero" is measured against the noise — a pixel whose fitted mean a0 = S0/2 is at most POLAR_DARK_K (3) × the noise of the modulation amplitude (σ̂·√(v₁+v₂), one σ̂ for the image from the residuals of an N > 3 sweep) is returned as 0.0 too, i.e. where the DoLP's own noise would be 1/3 or more. DoLP is a ratio, and clipped sensor noise in a dark region (σ 0.001 of a scene peaking at 1) used to come back as DoLP 1.0 — "completely polarised", the opposite of the truth — on 72 % of the dark pixels once *max_violation_frac* let the image through. A 0 here means "not defined", not "unpolarised" — use polarization_stokes_map's lit to tell them apart. With three angles there is no residual and only S0 = 0 is 0 (noise-only dark pixels are not screened). One σ̂ assumes stationary noise; shot noise grows with the signal, so the floor is set by the typical pixel.
Raises ValueError: as polarization_separate.
• specular_photometric 패밀리 가이드
• 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.
• example_polarization_metal — py -3.11 examples/example_polarization_metal.py
• perf_polarization_analyze — py -3.11 examples/perf_polarization_analyze.py
• poc_polarization_specular — py -3.11 examples/poc_polarization_specular.py
• polarization_camera_pipeline — py -3.11 examples/polarization_camera_pipeline.py
• specular_photometric — py -3.11 examples/specular_photometric.py
image2d 를 입력으로 받는 것)polarization_render · deflectometry_render · deflectometry_phase_to_slope
polarization)polarization_render · polarization_separate · polarization_stokes · polarization_stokes_map
*Provenance: specularity.py — SPECULAR 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.