rgb2lab([1,1,1]) returns [0.,0.,0.]

[jTiKey]

Description:

For some reason rgb2lab([1,1,1]) returns [0.,0.,0.] instead of array([ 1.00000000e+02, -2.45493786e-03, 4.65342115e-03]).
It works fine on my windows machine but the DO app doesn't work properly.

Way to reproduce:

from skimage.color import rgb2lab
rgb2lab([1,1,1])

Version information:

3.11.8 (main, Feb  7 2024, 09:33:13) [GCC 11.4.0]
Linux-4.4.0-x86_64-with-glibc2.35 (Digital ocean App platform)
scikit-image version: 0.23.2
numpy version: 1.26.4

[jTiKey]

After some tries, looks like the interger is the issue, replacing the integer with float makes it work.

[mkcor]

Thanks for using scikit-image, @jTiKey!

This is a matter of image data type, indeed: rgb2lab calls rgb2xyz which calls _prepare_colorarray which calls img_as_float32 or (in this case) img_as_float64 to convert the input array into floating-point representation... and:

import numpy as np
import skimage as ski

print(np.__version__)
1.26.4
print(ski.__version__)
0.23.2

arr = [1, 1, 1]
arr = np.asanyarray(arr)
print(arr.dtype)
int64

# Range of int64 goes up to 1/(2**64 - 1) so:
assert ski.util.dtype.img_as_float64(arr)[0] == 1/(2**64 - 1) * 2

See https://scikit-image.org/docs/stable/user_guide/data_types.html and references therein. Do you mind if I close this issue?