[BUG]: Conversion graph wrong results when converting between Lab and LCHab

[lassefschmidt]

Description

Description

I get weird results when using conversion graph between Lab and LCHab.

When I use as input the spectral distribution, conversion result is correct for both LCHab and Lab. But when I convert between Lab and LCHab, I get wrong results. In the backend, we use LCHab_to_Lab/Lab_to_LCHab. These functions are wrappers only and use JChab_to_Jab/Jab_to_JChab. But they have different annotations compared to LCHab_to_Lab -- maybe this causes the issue ?

Best,
Lasse

Code for Reproduction

import colour
import numpy as np

remission_curve = np.array([
    0.41408, 0.6332, 0.76334, 0.80015, 0.81076, 0.81892, 0.82485, 0.82879, 0.8314, 0.83452,
    0.83614, 0.83643, 0.83471, 0.83307, 0.83233, 0.82874, 0.82496, 0.82656, 0.82918, 0.82945,
    0.82993, 0.83085, 0.83332, 0.8376, 0.84379, 0.85061, 0.86037, 0.87132, 0.88176, 0.89274, 0.90209
])

wavelenghts = np.arange(400, 701, 10)

data = {
    float(wl): float(v) for wl, v in zip(wavelenghts, remission_curve)
}

reference = colour.SpectralDistribution(
    data
).align(colour.SpectralShape(400, 700, 1))

cmfs = colour.MSDS_CMFS['CIE 1964 10 Degree Standard Observer']
illuminant_sds = colour.SDS_ILLUMINANTS["D65"]
illuminant_ccs = colour.CCS_ILLUMINANTS['CIE 1964 10 Degree Standard Observer']["D65"]

lab = colour.convert(
    reference, "Spectral Distribution", "CIE Lab", to_reference_scale=True,
    sd_to_XYZ={"cmfs": cmfs, "illuminant": illuminant_sds, "method": "Integration"},
    XYZ_to_Lab={"illuminant": illuminant_ccs},
)

lch = colour.convert(
    reference, "Spectral Distribution", "CIE LCHab", to_reference_scale=True,
    sd_to_XYZ={"cmfs": cmfs, "illuminant": illuminant_sds, "method": "Integration"},
    XYZ_to_Lab={"illuminant": illuminant_ccs},
)

# these printouts are the correct values ! (verified)
print(lab, lch)

reconv_lab = colour.convert(lch, "CIE LCHab", "CIE Lab")
reconv_lch = colour.convert(lab, "CIE Lab", "CIE LCHab")

# these printouts are wrong (compare the values to previous printout -- the edges used in the conversion graph are LCHab_to_Lab [using JChab_to_Jab in backend which has other annotations compared to LCHab_to_Lab] and Lab_to_LCHab [same issue in backend])
print(reconv_lab, reconv_lab==lab)
print(reconv_lch, reconv_lch==lch)

Exception Message

Environment Information

===============================================================================
*                                                                             *
*   Interpreter :                                                             *
*       python : 3.14.0 (main, Oct 10 2025, 12:47:49) [Clang 20.1.4 ]         *
*                                                                             *
*   colour-science.org :                                                      *
*       colour : 0.4.7                                                        *
*                                                                             *
*   Runtime :                                                                 *
*       matplotlib : 3.10.7                                                   *
*       networkx : 3.5                                                        *
*       numpy : 2.3.3                                                         *
*                                                                             *
===============================================================================
defaultdict(dict,
            {'Interpreter': {'python': '3.14.0 (main, Oct 10 2025, 12:47:49) [Clang 20.1.4 ]'},
             'colour-science.org': {'colour': '0.4.7'},
             'Runtime': {'matplotlib': '3.10.7',
              'networkx': '3.5',
              'numpy': '2.3.3'}})

[KelSolaar]

I think that the wrappers under the new system are not producing full annotations. Looking!

[KelSolaar]

Pushed a fix in develop, that should hopefully be good!

[lschmidt-mkwz]

Another bug. This time a scaling issue in XYZ_to_Hunter_Lab (inputs for this function assumed to be in Domain100, and conversion graph does not correctly scale input before using it)

See below

import colour
import numpy as np

remission_curve = np.array([0.100000001, 0.100000001, 0.100000001, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0])

wavelenghts = np.arange(400, 701, 10)

data = {
    float(wl): float(v) for wl, v in zip(wavelenghts, remission_curve)
}

reference = colour.SpectralDistribution(
    data
).align(colour.SpectralShape(400, 700, 1))

cmfs = colour.MSDS_CMFS['CIE 1964 10 Degree Standard Observer']
illuminant_sds = colour.SDS_ILLUMINANTS["D65"]
illuminant_ccs = colour.CCS_ILLUMINANTS['CIE 1964 10 Degree Standard Observer']["D65"]
D65 = colour.TVS_ILLUMINANTS_HUNTERLAB["CIE 1964 10 Degree Standard Observer"]["D65"]

# get XYZ values (Domain1)
XYZ = colour.convert(
    reference, "Spectral Distribution", "CIE XYZ", to_reference_scale=False,
    sd_to_XYZ={"cmfs": cmfs, "illuminant": illuminant_sds, "method": "Integration"},
)

# from XYZ to hunter lab
print("This is the correct Hunter Lab value !")
print(colour.XYZ_to_Hunter_Lab(XYZ*100, D65.XYZ_n, D65.K_ab)) # input must be scale 100 (this is likely the issue ! We assume inputs to be Domain100 but conversion graph uses it with input Domain1)
print("=======")

# same but using conversion graph
print("This is wrong ! (should be in Domain100 but is in Domain1)")
print(colour.convert(
    XYZ, # input in scale 1
    "CIE XYZ",
    "Hunter Lab",
    to_reference_scale=True, # output in scale 100
    XYZ_to_Hunter_Lab={
        "XYZ_n": D65.XYZ_n,
        "K_ab": D65.K_ab,
    },
))
print("=======")

# from sds to hunter lab
print("This is wrong ! (should be in Domain100 but is in Domain1)")
print(colour.convert(
    reference,
    "Spectral Distribution",
    "Hunter Lab",
    to_reference_scale=True,
    sd_to_XYZ={"cmfs": cmfs, "illuminant": illuminant_sds, "method": "Integration"},
    XYZ_to_Hunter_Lab={
        "XYZ_n": D65.XYZ_n,
        "K_ab": D65.K_ab,
    },
))
print("=======")

This returns the following prints

This is the correct Hunter Lab value !
[  1.59245061  24.93814936 -43.42988114]
=======
This is wrong ! (should be in Domain100 but is in Domain1)
[ 0.15924506  2.49381494 -4.34298811]
=======
This is wrong ! (should be in Domain100 but is in Domain1)
[ 0.15924506  2.49381494 -4.34298811]
=======

[KelSolaar]

Hi @lassefschmidt ,

I haven't tested but isn't the issue that you are passing XYZ_n and K_ab unscaled when your input is Scale 1?

print("This is wrong ! (should be in Domain100 but is in Domain1)")
print(colour.convert(
    XYZ, # input in scale 1
    "CIE XYZ",
    "Hunter Lab",
    to_reference_scale=True, # output in scale 100
    XYZ_to_Hunter_Lab={
        "XYZ_n": D65.XYZ_n, # <-- Should not it be scaled to Scale 1 manually (i.e., / 100)?
        "K_ab": D65.K_ab, # <-- Should not it be scaled to Scale 1 manually (i.e., / 100)?
    },
))

[lschmidt-mkwz]

Hi @lassefschmidt ,

I haven't tested but isn't the issue that you are passing XYZ_n and K_ab unscaled when your input is Scale 1?

print("This is wrong ! (should be in Domain100 but is in Domain1)")
print(colour.convert(
XYZ, # input in scale 1
"CIE XYZ",
"Hunter Lab",
to_reference_scale=True, # output in scale 100
XYZ_to_Hunter_Lab={
"XYZ_n": D65.XYZ_n, # <-- Should not it be scaled to Scale 1 manually (i.e., / 100)?
"K_ab": D65.K_ab, # <-- Should not it be scaled to Scale 1 manually (i.e., / 100)?
},
))

When using (dividing by 100)

XYZ_to_Hunter_Lab={
      "XYZ_n": illuminant_tvs_hunterlab.XYZ_n/100,
      "K_ab": illuminant_tvs_hunterlab.K_ab/100,
},

we get the following prints

This is the correct Hunter Lab value !
[  1.59245061  24.93814936 -43.42988114]
=======
This is still wrong
[ 1.59245061  0.24938149 -0.43429881]
=======
This is still wrong
[ 1.59245061  0.24938149 -0.43429881]
=======

I added the prints without the scaling to the previous message to easily compare !

To get the correct output everywhere, you would have to scale only XYZ_n, but I am unsure if this is intended behaviour.

XYZ_to_Hunter_Lab={
    "XYZ_n": illuminant_tvs_hunterlab.XYZ_n/100,
    "K_ab": illuminant_tvs_hunterlab.K_ab,
},

This would return for all prints
[ 1.59245061 24.93814936 -43.42988114]

On a general note : given that all internal operations in the conversion graph are in Domain1, shouldn't we automatically convert the passed illuminants to the right scale ? e.g. for CIE Lab, I never had to scale the desired colour.CCS_ILLUMINANTS (as it is in Domain1 for chromaticity already). And when you pass the Domain1 ccs_illuminant to Lab_to_XYZ it also works fine ! At least for me, your suggestion to scale XYZ_n and K_ab was very unintuitive as never before I had to scale the parameters. Maybe it would make sense (at some point) to also define the domain of these objects (such as TVS_ILLUMINANTS) ?

[KelSolaar]

Ah yes sorry, I should have been only scaling the illuminant in my example:K_ab is a chromaticity coefficient that is scale-independent.

This is intended behaviour, if you look at the docstring, we document the expected scale:

def XYZ_to_Hunter_Lab(
    XYZ: Domain100,
    XYZ_n: ArrayLike | None = None,
    K_ab: ArrayLike | None = None,
) -> Range100:
    """
    Convert from *CIE XYZ* tristimulus values to *Hunter L,a,b* colour
    
    ...

    Notes
    -----
    +------------+-----------------------+-----------------+
    | **Domain** | **Scale - Reference** | **Scale - 1**   |
    +============+=======================+=================+
    | ``XYZ``    | 100                   | 1               |
    +------------+-----------------------+-----------------+
    | ``XYZ_n``  | 100                   | 1               |
    +------------+-----------------------+-----------------+

    +------------+-----------------------+-----------------+
    | **Range**  | **Scale - Reference** | **Scale - 1**   |
    +============+=======================+=================+
    | ``Lab``    | 100                   | 1               |
    +------------+-----------------------+-----------------+

    ...

    """

For XYZ_to_Lab, no such problem because it is always Scale 1:

def XYZ_to_Lab(
    XYZ: Domain1,
    illuminant: ArrayLike = CCS_ILLUMINANTS["CIE 1931 2 Degree Standard Observer"][
        "D65"
    ],
) -> Range100:
    """
    Convert from *CIE XYZ* tristimulus values to *CIE L\\*a\\*b\\**
    
    ...

    Notes
    -----
    +----------------+-----------------------+-----------------+
    | **Domain**     | **Scale - Reference** | **Scale - 1**   |
    +================+=======================+=================+
    | ``XYZ``        | 1                     | 1               |
    +----------------+-----------------------+-----------------+
    | ``illuminant`` | 1                     | 1               |
    +----------------+-----------------------+-----------------+

    +----------------+-----------------------+-----------------+
    | **Range**      | **Scale - Reference** | **Scale - 1**   |
    +================+=======================+=================+
    | ``Lab``        | 100                   | 1               |
    +----------------+-----------------------+-----------------+

    ...

    """

The core difference is that the former function takes CIE XYZ tristimulus values and the other one CIE xy chromaticity coordinates for the illuminant.

[lschmidt-mkwz]

Thank you for the information ! I think this can be closed then. My bad, should have read better !