Why *a and *b are divided by 110?

[ZenithFlux]

I think in colorization_dataset.py, the ab tensor should be divided by 128, not 110, as scikit-image implementation of L*a*b has a range of -128 to 127 for *a and *b.

The range of ab is -110 to 110 in this Matlab implementation.So I think this is the source of 110. But i dont think this is valid for scikit, is it?

[junyanz]

This is a great question. Originally, we followed the Torch implementation of pix2pix. See this line. But I am not sure if the Lua image and skimage produce the same or different ab range.
I did some Google search regarding the Lab range of skimage, but could not find a definite answer. This post mentioned -128, 128, but this post mentioned a narrow range, which is within -110, 110.