Target and lighting

The printed board 𝓑 and its warping h, the Cook–Torrance SVBSDF and the radiance L_c. The light is VisualRadiometryToolkitCore's EnvironmentLight (real spherical harmonics per channel on a sphere about the board centre, SphericalHarmonicMap); its documentation is that package's.

The board and its warping

RadiometricCamera.HEIGHT_NFREE — Constant
HEIGHT_NFREE = HEIGHT_NC - 3

The number of free parameters of the board warping h: the pure-bending nodal heights (Core's tangent_size of the warping's TargetManifold, height_basis, the same on a page of any aspect). The piston and the two tilts are the poses', so they are not parameters of h.

source
RadiometricCamera.height_basis — Method
height_basis(b) -> SMatrix{HEIGHT_NC, HEIGHT_NFREE}

Core's retract_jacobian of height_manifold(b): a basis of the page's pure-bending nodal heights, h = h₀ + B y (the free height parameters y of the refinement's material block, MaterialVar), static for the kernels' dual arithmetic.

source
RadiometricCamera.height_manifold — Method
height_manifold(b) -> VgtkCore.TargetManifold

The manifold of the board b's nodal heights: Core's TargetManifold of a HeightField over its page's warping_nodes. Core's thin-plate kinds are pure bending and have no gauge: the nodal heights move only within the heights whose interpolant on these nodes has no affine term (the piston and the two tilts are the poses'), so they are never parameters. The affine rows depend on the aspect, so each page has its own manifold.

source
RadiometricCamera.height_nparams — Method

Number of free height-field parameters of degree 0 (none: the plane) or ≥ 1 (all HEIGHT_NC nodal heights; the three affine combinations — piston and the two tilts, the poses' — are null directions of the representation, so they cannot be tilted into).

source
RadiometricCamera.load_distance_map — Method
load_distance_map(path) -> (values::Matrix{Float32}, page_mm::NTuple{2, Float64})

The printed design's edge distance map (edge.sdf): a VisualGeometryToolkitIO raster record (vgio/raster/v2.0, load_raster: float32, the page chart's rectangle, row 1 its top edge) — its values (rows × columns, mm, + on paper, − in ink) and the page's size in mm, the chart's widths. The one reader (owner D6, 2026-11-15: no hand parser; the generator's raw sdf.f32 is converted by the experiments' scripts/distance_maps.jl, a keyed step). A raster in another frame, or one starting off the page's corner, is refused by name.

source
RadiometricCamera.sphere_radius — Method
sphere_radius(𝔟, centres) -> r_SH

The light-sphere radius max_k ‖𝐜_k − 𝔟‖ about the point 𝔟 from the camera centres centres (the radius of Core's EnvironmentLight about 𝔟).

source
RadiometricCamera.sphere_radius — Method
sphere_radius(b::TexturedTarget, centres) -> r_SH

The pipeline's light-sphere radius about the board centre 𝔟: the farthest of the camera centres AND of the page's corners (sphere_radius(𝔟(b), centres) is the cameras alone). Every board point then lies inside the sphere, so every light ray from the page leaves it through the sphere (Core's exit_direction with Δ ≥ 0) — with cameras closer than the page's half-diagonal (the 27.10 set: 0.61 m against 0.71 m) the corners lay outside, where the exit falls back to the ray's closest point (2026-11-04).

source
RadiometricCamera.RadiometricTarget — Type
RadiometricTarget(target, warping, aspect, radiometry)

A calibration target as a refinement knows it, parallel to the geometric camera ↔ the RadiometricCameraModel: the geometric target (target, Core's AbstractTarget as the session's target record holds it, so its content identity is the session's; nothing for a state converted from a RadiometricCamera 0.5 .jls file, which did not record it), the board's warping as the refinement learned it (warping, the HEIGHT_NC nodal heights of TargetState.h, plain metres along the board normal — a calibration of the target that its content identity does not cover, so not the target's own HeightField), the aspect (height / width) of the page the warping was learned on (aspect, warping_aspect of its board: the same nodal heights are another surface on a page of another aspect), the in-plane print error as the refinement learned it (inplane, the (a, u) of Core's Aspect ∘ Skew in the warping's placed chart — inplane_calibration; zero = the design as printed) and the print's radiometry (TargetRadiometry). A flat warping (every nodal height zero) is the plane on any page: its aspect may be nothing; a warped one, or an in-plane error (whose chart depends on the aspect), without an aspect is an ArgumentError. board_surface is the board's TargetState of it on a board.

source
RadiometricCamera.TargetRadiometry — Type
TargetRadiometry(material, ink_spread, print_blur)
TargetRadiometry(material, surface::TargetState)

The radiometry of a printed calibration target: its SVBSDF (material, a Material: Θm = the roughnesses and refractive indices of ink and paper and the diffuse weight μ) and, for a design given as a distance map, its printed edge — the ink spread δ (`inkspread, a length: a FlexUnits quantity in metres) and the print's own blur (printblur, Core'sGaussianBluron the print, round, its moment in m²); a plain ink spread or a plain (px²) print blur is anArgumentError. From a state's working form,materialand the surface'sedge = (δ, σprint)in mm, read decimal-exactly (VisualGeometryToolkitIO'sreadlength):edgeparameters(r)` gives the mm values back bit for bit for values of up to 15 significant digits.

source
RadiometricCamera.board_surface — Method
board_surface(b::TexturedTarget, t::RadiometricTarget) -> TargetState
board_surface(aspect, t::RadiometricTarget) -> TargetState

The TargetState of a radiometric target on the board b (or a page of aspect = height / width): its warping, its in-plane print error and its print's edge parameters. The target's warping and in-plane error must have been learned on a page of the board's aspect (t.aspect == warping_aspect(b) to 1e-6), or be flat and as designed: an ArgumentError otherwise, since the nodal heights (and the placed chart of the in-plane error) on a page of another aspect are another surface.

source

SVBSDF and radiance

RadiometricCamera.RadianceParts — Type
RadianceParts{J, T}

Bilinear decomposition of the estimator: L_c = Σ_j a[c,j] (μ Bd[c,j] + (1−μ) Bs[c,j]), i.e. Bd/Bs are the diffuse/specular response coefficients of the SH basis (both linear in Θ_l, the combination affine in μ).

source
RadiometricCamera.RadianceSampler — Type
RadianceSampler

N VNDF samples per (𝐱, ω); seed of the counter-based random stream; hammersley uses the deterministic Hammersley set (scrambled per key) instead of pseudo-random draws.

source
RadiometricCamera.L_c — Method
L_c(board, mat, light, sampler, key, 𝐛, 𝐱, ω) -> SVector{3}

Monte-Carlo estimate of the per-channel radiance of the point 𝐱 (chart 𝐛) of the board (a TexturedTarget, under its state: its printed edge and in-plane print error) seen along ω, the unit direction from the aperture point to the board point (world frame; the paper's π̂⁻¹(𝐩, 𝐚) − 𝐚). The radiance leaves the board towards −ω.

source
RadiometricCamera.draws — Method
draws(sampler, [::Type{T} = Float64,] h_or_key, i) -> (u₁, u₂)

Two T-typed uniforms in (0, 1) for sample i of a key hash h (counter-based, stateless). The type follows the sampler (Julia's order, as rand(rng, T, dims): the sampler is the random stream).

source
RadiometricCamera.radiance_parts — Method
radiance_parts(board, mat, light, sampler, key, 𝐛, 𝐱, ω) -> RadianceParts

Same samples and direction convention as L_c (ω from the aperture point to the board point); used to build the design matrices of the linear solves.

source