Stacked on rb/1592: LOOK_VERSION 8 -> 9; the PR's corpus-profile helper is the base's corpus_uses_the_bundled_profile.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Resolved against rb-base-36: LOOK_VERSION 6 -> 7; starting_colour_tags passes no lens (merge frames arrive without it).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Move nested conditions into match guards and simplify equivalent Boolean
expressions without changing validation or short-circuit behavior. Keep
the explicit permitted Testkit dev-dependency arm in the layer checker.
Co-Authored-By: OpenAI Codex <noreply@openai.com>
Brings in batch 35 / v0.8.0. Code merged cleanly; conflicts only in docs: the LR-MERGE-HDR row keeps main's
lens-correction text plus this branch's white-balance item, and this branch's What's New entry moves to the
2026-10-14 section.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015gRtdWF1E9358FfzhzYqCS
Sony ARWs record shading, lateral chromatic aberration and distortion as a camera setting plus a radial
table in the raw image IFD (0x7031-0x7037), with copies in the encrypted SR2SubIFD (0x797c-0x7982).
They are now decoded for every body that records them and carried through RawInfo and RawImage as
`lens_metadata.sony` (types in crates/raw/src/lens_metadata.rs, decoder in
crates/raw/src/vendor/arw_lens.rs):
- Raw-IFD fields are the ones the TIFF reader parsed, which is what the image itself uses. A scan of
the directory adds diagnostics only: repeated tags, and why an entry the reader skipped was dropped
(values outside the file, an unknown type, past the reader's limits). SR2SubIFD fields are read
from their first entry. Only the exact layouts are accepted: SHORT[1] or non-negative SSHORT[1]
settings, SSHORT[17] headed by 16 for shading and distortion, SSHORT[33] headed by 32 for CA.
Nothing is allocated from a count in the file.
- The raw IFD's copy wins. The SR2SubIFD copy fills in only for a tag the raw IFD lacks, never for a
malformed one, and the two copies' agreement is recorded.
- Settings keep their number and their documented meaning, Unknown(value) otherwise (never Off).
- Framing evidence (stored frame, Sony and default crops, reference rectangle, coding, 3:2) and a
per-component status. Only Applied components become opcodes.
Applied corrections are unchanged: the ILCE-7RM4A distortion conversion receives exactly the table it
received before, and no other body or component gains a correction. The ILCE-7CR distortion misses its
camera JPEG at 24 mm, and shading and CA semantics are not validated on any body; see
docs/sony-lens-corrections.md.
Locating the SR2SubIFD, now done for every ARW that has one, selects the SR2Private offset, length and
key with the TIFF reader's own rules, which lightcraft-tiff now exposes and builds its reader on
(directory_entries, entry_span, ParseOptions::budget_for, decode_value). Only each field's first value
is decoded, so the directory's other entries cost no memory sized by the file: a 16 MiB file whose
SR2Private holds eight fields aliasing it peaks at 37 kB. DirectoryEntries keeps its layout private,
and its positions() iterator uses checked arithmetic.
Tests: synthetic decoder tests; a differential test of the SR2Private selection against the TIFF
reader's in both byte orders (0 to 8,193 filler entries, unreadable and unusual first occurrences of
each field, a used-up value budget); a dhat peak-heap test; public-corpus checks against the values
ExifTool prints, with pinned and anchored SHA-256s (LIGHTCRAFT_CORPUS_REQUIRED=1 makes absent files
fail); and an opt-in local inventory check (crates/raw/tests/sony_lens_inventory.rs).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Fitted with `lightcraft-cli calibrate` to 144 usable ILCA-99M2 ARWs (of
150, seven shooting days, Standard creative style, DRO Auto) and their
in-camera JPEGs; aggregate colour statistics only.
On 30 withheld photos (profile fitted to the other 120) mean dE76 against
the camera JPEG goes 5.11 -> 4.87; 25 of 30 improve, the worst
regression is 0.24. All 150 photos stay accepted. The public raw.pixls.us
ILCA-99M2 file, not used for fitting, stays accepted (5.57 -> 5.73).
A corpus test checks that this public sample's fit keeps exactly the
built-in profile's matrix and table; the ILCE-7CR test now shares its
body. LOOK_VERSION 5 -> 6, as the profile changes the fit of this model.
`embedded_preview_srgb` (the loupe stand-in and the grid thumbnail of an unedited raw) decoded the
largest embedded JPEG whatever the size asked for: a 512 px thumbnail of a Sony A7R VI ARW decoded
its 9984 x 6656 JPEG (5.6 MB) although the file holds a 1616 x 1080 one (0.3 MB).
`lightcraft_raw::embedded_preview_covering(bytes, min_edge)` returns the smallest of a TIFF-based
raw's JPEGs whose long edge is at least `min_edge` and that is a copy of the largest one: aspect
ratio within 1 % (not a letterboxed thumbnail of another shape) and the same number of components
(not the grey semantic masks of an iPhone ProRAW DNG); else the largest, as `embedded_preview`.
The colour-space handling stays in `decode_raw_preview` (a `covering` argument); an unsupported
raw's editing source and the camera-look fit keep the largest JPEG, so a source never depends on
its size level.
On 120 raws of every make in the corpus at 256 / 512 / 1600 px: wherever the largest JPEG is still
chosen the output is identical byte for byte; where a smaller one is (Sony, Nikon, Panasonic,
Adobe DNG) orientation matches, the size is equal or 1-2 px off, mean difference <= 4.7 / 255.
- Rebased onto #1020: its plan route (`preview_window` -> `planned_preview`) now takes the box size and picks
with `embedded_preview_covering` too, so a TIFF raw read through the sparse plan gets the whole file's JPEG.
Co-Authored-By: Claude <noreply@anthropic.com>
The hue/saturation table of a fitted camera look (ILCE-7RM4A, APS-C ILCE-7RM6, ...) ran
HsvTable::apply pixel by pixel: branches on the data (largest channel, hue sector),
saturating float-to-usize casts and bounds-checked table reads, ~550 instructions and 1.2
branch misses per pixel. CameraColourModel::apply now runs it 16 pixels at a time
(HueSat::apply_lanes -> HsvTable::apply_lanes, structure of arrays): per pixel the same IEEE
operations in the same order, with the branches turned into selects and the table indices
computed as exact f32 integers, so the compiler vectorises everything but the table reads,
powf and the rare rem_euclid, which run lane by lane. DNG profile tables and the other colour
models keep the per-pixel path.
Measured on a Core Ultra 9 185H P-core (busy machine, min of N):
- HsvTable alone, 72x5x5 fitted-look-shaped table, random colours, one thread:
243 -> 106 cycles (57 -> 26 ns) per pixel.
- Colour stage of a full-size load, one thread: 60 MP ILCE-7RM4A 3.80 -> 1.99 s,
26 MP APS-C ILCE-7RM6 HQ 1.88 -> 0.98 s; 4 CPUs: 1.02 -> 0.67 s, 0.56 -> 0.30 s.
Exactness: new tests compare the lanes with the one-by-one path bit for bit (to_bits) on tens
of thousands of colours (every combination of zero, -0, tiny, negative, around 1, huge, NaN and
+-inf components; grey; hues at the 0/360 degree wrap; near-black noise in the restore's blend)
for fitted, value-scaled, 2D, many-turn and degenerate tables, 16 and 5 lanes, and on an image
whose rows aren't a whole number of lanes. Stage fingerprints (decode, AHD, 2560 px and
full-size loads) of 7 Sony raws are identical to origin/main.
Sources: own design.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
The D1X/D1H raw SubIFD is 4028x1324 with XResolution 600 / YResolution 300:
the pixel is twice as tall as it is wide. Nothing read that, so the
picture came out about 3:1. Add a pixel aspect to RawImage/RawInfo, read
it from the raw IFD resolution ratio (NEF) and from DefaultScale (DNG),
accept only 1.5..4, and resample the rows after demosaic. developed_size()
reports the stretched size, binning is off for non-square pixels, and the
catalog and lens-profile sizing use developed_size().
A raw whose white balance is only in the vendor's multipliers (CR2, NEF, ARW...)
has no AsShotNeutral of its own. Loading a frame dropped the multipliers and
stitch copied the first frame's colour tags, so a panorama of frames that reach
the stitcher without a neutral wrote a DNG with neither AsShotNeutral nor
AsShotWhiteXY, and a viewer had no white to start from (a strong colour cast
until White Balance > Auto).
Frames now carry the multipliers (Frame::wb_multipliers, filled from the decoded
raw; load_frame_with is split so a decoded raw can be tested without a file).
stitch takes the median R/G and B/G of the multipliers of the frames that are in
the panorama (the mean of the middle two for an even count; frames without
multipliers, or with non-finite or non-positive ones, do not vote) and sets the
result's AsShotNeutral to their reciprocal, green = 1. The source's
AsShotWhiteXY is dropped with it: a reader takes it before the neutral and the
DNG spec makes the two exclusive. With no multipliers anywhere (DNG sources) the
first frame's tags are kept, as before.
In the app and the CLI a raw without colour matrices already gets a neutral from
starting_colour_tags (#965). For frames balanced alike the new neutral renders
the same (checked in the engine crate); it differs when the camera balanced the
frames differently: the panorama follows the median instead of the first frame.
Tests: the frame carries the multipliers; the consensus for odd and even counts,
with missing and unusable values; stitching raw-style frames gives the median
AsShotNeutral, ignores frames outside the panorama and leaves the colour as it
was without multipliers; the written DNG, read back with the raw reader, has the
reciprocal AsShotNeutral and the white balance the source's multipliers give.
#1123 gives HDR and panorama merges of raws without colour matrices the colour they open with, but writes
ColorMatrix1 from the forward matrix, so the as-shot white the DNG declares lands wherever the fitted look and the
generic camera model take it. On real ILCE-7M3 brackets the HDRs opened at As Shot 6113–7307 K / +115 (Duv +0.038)
where the ARWs show 3932–5268 K / +7, and any white-balance edit turned them green (CIEDE2000 11–13 against the ARW
at the same Temp/Tint). The render at As Shot was right (0.4–0.8).
The colour matrices now declare the white balance the source shows, decided by files::WbScaleOf (new dng_white):
the camera's measured matrices when they give the source's scale (lightcraft_raw::spectral; the DNG then reads every
white as the source does), else a matrix under which the as-shot neutral is the white the scale shows (Sony's
presets' reading, 6500 K / 0 on the relative scale, the file's own white otherwise). Forward matrix and neutral are
unchanged, so is the render at As Shot.
Measured on the same brackets with this change on a fork build: As Shot 4813 / 5268 / 3932 K, +8 / +8 / +7 as the
ARWs (declared white Duv +0.003); same Temp/Tint within CIEDE2000 0.8–1.7 of the ARW for most changes, about 5 for
the largest (5268 → 3200 K, 3932 → 6500 K: the ARW adapts its look with Bradford, the DNG rebalances in camera space).
Tests: a merged DNG shows its source's As Shot on all four scales (measured matrices, presets, relative, own); with a
look that tints white the DNG declares the shown white, gives the measured model's multipliers at other whites and
renders at As Shot as the source does. Both fail on main (ILCE-7M3 case: (7342, 115) for (5268, 8)).
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015gRtdWF1E9358FfzhzYqCS
Sony ARWs of cameras without measured colour matrices were edited on the
relative scale (6500 K / 0 = as shot). Their SR2SubIFD stores the
multipliers of the camera's presets at 2500, 3200, 4500, 6000 and 8500 K;
lightcraft_raw::kelvin_presets reads the as-shot multipliers on the curve
through them and they take the spectral matrices' place in the #730
design (as-shot Kelvin and tint only; look, Bradford adaptation and the
pixel-identical legacy upgrade unchanged).
Covered: ILCA-99M2, ILCA-77M2, SLT-A33, DSC-RX10, ILCE-3500, ILCE-7S.
Kelvin-dial shots read back 3400 -> 3396, 5200 -> 5193, 4700 -> 4708,
5600 -> 5594 K; Shade 7493 K. Off-curve constants (isotherm tilt 0.74,
tint spacing 1) measured from the 14 Sony bodies with both spectral
matrices and presets.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Add correlated dark holdouts that reject misleading training and accept
matching training. Check finite, monotone curves in both cases. Confirm
the regression fails with the error gate removed and passes when restored.
Qualify the improvement claim when no original curve can be fitted, and
explain the source-look process policy and offline smart preview refresh.
Co-Authored-By: Codex <noreply@openai.com>
Use separate luminance pairs for tone fitting so the camera look follows
its JPEG into deep shadows. Keep colour pairs and holdouts unchanged.
Require enough correlated shadow samples and lower held-out error, with
noise and compression floors and the existing median curve fit.
Add synthetic regressions, bump look and render cache versions, and
record the public multi-make and private numeric validation in the docs.
Fixes#1307
Co-Authored-By: Codex <noreply@openai.com>
For a DNG with its own colour matrix, an identity or absent ProfileToneCurve, no
LookTable and no ProfileGainTableMap, fit only the tone and chroma curves to the
camera JPEG, with the file's colour model untouched (profile tables and embedded
vignetting applied to the sensor proxy as the loader renders it). The fit is used
only when the preview is at least 90% of the raw's long edge, shows the same
picture (structure gate), passes the existing held-out acceptance, and does not
put more of the photo near white than the plain render (highlight gate, at 2x
proxy luminance). Otherwise the render is unchanged. Other formats and
catalogued photos are untouched. Render cache version 62.
libtest has no skipped status and hides the output of passing tests, so a
test that returns early without a GPU, a corpus file, a craft-fonts face or
a model passed silently. `lightcraft-testkit` (the testkit slot the layering
table already reserves, dev-dependency only, std only) adds
`skip!(Need, "reason")`: it prints the note as before and, when
LIGHTCRAFT_SKIP_LOG names a file, appends one JSON line (need, reason, test,
binary, file:line). Plain `cargo test` is unchanged. The 61 existing skip
sites use it.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Photo Merge decoded each raw itself and wrote the reference frame's own colour
tags into the merged DNG. A CR2 (and NEF, ARW and the other raws whose colour is
fitted from the camera JPEG) has none, so an HDR merge of CR2s rendered with no
camera colour: green and about 16 L* dark.
`lightcraft_merge::load_frame_with` takes the raw's colour tags from the caller,
and the engine passes `starting_colour_tags`: the colour the raw opens with
(`starting_colour`) as DNG tags. That is one forward matrix with a colour matrix
and AsShotNeutral that agree with it at the as-shot white, the look's hue/sat
table as ProfileHueSatMap and its tone curve as ProfileToneCurve. Raws with
their own matrices (DNG) keep their tags. The look's chroma curve has no DNG
tag and is left out.
On two CC0 CR2s (EOS 550D, 40D) an HDR merge of three copies against the photo
alone: mean CIE76 dE 26.15 -> 1.07 and 22.89 -> 0.47.
Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
