Numa's own camera profiles, v1

Numa's own work. Each profile was fitted by Numa (tests/camera_profile_fit.rs,
RENDER-012/015) from raw files that raw.pixls.us publishes under CC0 1.0
(public domain), against the JPEG each camera embedded in its own file. No
other camera profile, and no photograph under any other licence, went into
them. They are offered beside a camera's matrix and any other profile, and
since RENDER-016 Automatic takes Numa's own first where a body has one
(Preferences, General, Colour, Camera profile can choose otherwise).

Each is shipped only because it beat the camera's matrix on held-out frames
(frames the fit never saw), as mean chroma error against the camera's JPEG —
refitted on 27 September on the cameras' tone curve (RENDER-004), on which the
matrix alone is already nearer the JPEGs than it was. The E-M1 Mark II's is
gone: fitted then against Numa's own rendering, since rawler reads no preview
from an ORF, and fitted again against the camera's it did not beat the matrix.

The five fitted from raw.pixls.us beside the X-T5 on 27 September (EOS R5,
D850, ILCE-7RM3, X-S10, X-T4) are gone since 29 September: each was held out on a frame of the same
session it was fitted on. The ILCE-7RM3's, measured on a frame of another scene,
did not beat the matrix; the other four have no second scene to measure on. The
X-T5's, measured on two of the photographer's later frames and three CC0 ones,
still does (0.0375 -> 0.0308, Adobe Standard 0.0372).

The twenty beside the X-T5 were added on 29 September (RENDER-023), from up to
four raws a body: each beat the matrix and, where one is installed, the
RawTherapee profile, on a held-out frame of a scene the fit never saw (taken
more than half an hour from every fitted frame). 220 bodies were fitted; most
of the rest were held back because every raw of the body shows one scene, so
no frame could be held out honestly.

Numa X-T5.dcp            Fujifilm X-T5           0.0228 -> 0.0160 (3 held out)  the photographer's own, see below
Numa ILCE-6300.dcp       Sony ILCE-6300          0.0260 -> 0.0258 (1 held out)  raws 783 979
Numa ILCE-6400A.dcp      Sony ILCE-6400A         0.0839 -> 0.0761 (1 held out)  raws 7826 7895
Numa ILCE-7M3.dcp        Sony ILCE-7M3           0.0333 -> 0.0271 (1 held out)  raws 2414 2415 2418
Numa EOS 850D.dcp        Canon EOS 850D          0.0618 -> 0.0600 (1 held out)  raws 4867 4868
Numa EOS M50.dcp         Canon EOS M50           0.0248 -> 0.0223 (1 held out)  raws 2663 4656 4657
Numa EOS R.dcp           Canon EOS R             0.0239 -> 0.0133 (1 held out)  raws 4515 4543 4643 4668
Numa 1 V1.dcp            Nikon 1 V1              0.0267 -> 0.0186 (1 held out)  raws 2210 2217
Numa D200.dcp            Nikon D200              0.0413 -> 0.0345 (1 held out)  raws 8079 824
Numa D3400.dcp           Nikon D3400             0.0349 -> 0.0291 (1 held out)  raws 1561 1876
Numa D5000.dcp           Nikon D5000             0.0639 -> 0.0617 (1 held out)  raws 1341 2825
Numa Z fc.dcp            Nikon Z fc              0.0753 -> 0.0708 (1 held out)  raws 4812 4813 8563 8566
Numa FC7303.dcp          DJI FC7303              0.0700 -> 0.0613 (1 held out)  raws 4785 4898
Numa C (Typ 112).dcp     Leica C (Typ 112)       0.0154 -> 0.0148 (1 held out)  raws 3017 4963
Numa E-M10 Mark IV.dcp   Olympus E-M10 Mark IV   0.0289 -> 0.0270 (1 held out)  raws 4126 4132
Numa DMC-LX3.dcp         Panasonic DMC-LX3       0.0759 -> 0.0635 (1 held out)  raws 1395 1808
Numa DMC-TX1.dcp         Panasonic DMC-TX1       0.0557 -> 0.0409 (1 held out)  raws 3103 3105 3106 3107
Numa K-5 II s.dcp        Pentax K-5 II s         0.0275 -> 0.0249 (1 held out)  raws 1184 1198
Numa K-70.dcp            Pentax K-70             0.0817 -> 0.0563 (1 held out)  raws 1141 1142 1753
Numa K-S1.dcp            Pentax K-S1             0.0830 -> 0.0615 (1 held out)  raws 1365 4365
Numa K20D.dcp            Pentax K20D             0.0421 -> 0.0339 (1 held out)  raws 1143 2879

The X-T5 profile is the one exception to the raw.pixls.us rule: it was fitted
from ten of the photographer's own X-T5 frames (Provia, Color 0) against the
JPEG each carries, three of them held out, on the cameras' tone curve
(RENDER-004, 27 September). Against the camera's HEIF of the same frames the
median colour difference (CIEDE2000) goes from 1.86 to 1.07, where Capture One's
Provia is 1.06 and Lightroom's Adobe Color 1.47.

The raws: https://raw.pixls.us/getfile.php/<number>/nice/ (CC0 1.0,
https://creativecommons.org/publicdomain/zero/1.0/).

© 2026 Tijmen. All rights reserved. Included with Numa; not for separate
redistribution. (The CC0 raws above are the raw.pixls.us contributors'; the
profiles fitted from them are not CC0.)
