How to convert to float? / add standard convert methods?

[tbenst]

I don't know how to convert a RGB to a tuple of UInts/Ints/floats, and it's surprisingly hard to figure this out. I don't see it documented, and when I try to find pertinent methods: methodswith(typeof(RGB{N0f8}(100/255,2/255,20/255)), supertypes=true), I get promising hits like float64 and these convert methods:

[85] convert(::Type{C}, c::Color, alpha) where C<:TransparentColor in ColorTypes at /home/tyler/.julia/packages/ColorTypes/RF8lb/src/conversions.jl:78
[86] convert(::Type{C}, c::C) where C<:Colorant in ColorTypes at /home/tyler/.julia/packages/ColorTypes/RF8lb/src/conversions.jl:72
[87] convert(::Type{C}, c::Colorant) where C<:Colorant in ColorTypes at /home/tyler/.julia/packages/ColorTypes/RF8lb/src/conversions.jl:73

float64 does not return primitive values for example. The closest I've figured out is reinterpret(UInt8, gray(Gray(RGB{N0f8}(100/255,2/255,20/255)))) but that of course isn't all three channels and seems very hard for such a simple task.

Any tips? And perhaps this should be a feature request too for these convert methods to be added

[timholy]

Are you using ImageCore? channelview and colorview are the standard methods for arrays. It also has rawview and normedview. Docs are at https://juliaimages.org/latest/.

The topic is a bit subtle because we support serveral different representations of AbstractRGB: RGB, BGR, XRGB and RGBX (two ways of padding up to 32 bits, good for GPUs), and RGB24. reinterpret takes the bitlevel representation at face value (i.e., storage-order) but constructors decouple representation from meaning. For a single c::AbstractRGB, the safest approach is just (red(c), green(c), blue(c)) because then you're in perfect control of what you get out. See https://github.com/JuliaGraphics/ColorTypes.jl#traits-utility-functions-for-instances-and-types.

[timholy]

Any further thoughts on this topic @tbenst? Can we close this or is there a specific suggestion for improving the documentation?

[tbenst]

thank you! channelview is close to what I was looking for, but alas this does not work: channelview(RGB{N0f8}(100/255,2/255,20/255)), however channelview([RGB{N0f8}(100/255,2/255,20/255)]) does work. Perhaps it's worth highlighting channelview on the Colors docs? It is indeed nicely documented in ImageCore, although I wonder if it's worthwhile to implement a method for convert that just calls channelview, maybe with a copy since perhaps other beginners might expect this to exist?

[timholy]

In Julia, it's common that anything called "view" references an AbstractArray.

If you're working with an immutable value (like a single RGB), it's extremely wasteful to first create an array out of it just to do color channel manipulations. Can you describe for me more clearly why you are trying to do it this way? What's bad about

c = RGB{N0f8}(100/255,2/255,20/255)
t = (red(c), green(c), blue(c))   # tuple form

[tbenst]

The tuple form makes sense! That’s what I was originally looking for—how to get the plain old float values out of RGB. I’m quite used to the opencv/numpy/scikit-image way of doing things and got stuck when summing an image, expecting a float, but getting an RGB.

It’s great that these functions exist, but I’m not sure how I would have discovered the functions red, green and blue without you pointing me to them. Perhaps these could be listed on Construction and Conversion? http://juliagraphics.github.io/Colors.jl/stable/constructionandconversion/

[timholy]

I linked them in #447 (comment). Maybe the problem is the split between ColorTypes and Colors?

[tbenst]

Ah indeed, that documents this quite clearly. I see it is also linked at http://juliagraphics.github.io/Colors.jl/stable/#Reexport-1. Maybe a reciprocal link to ColorTypes on this repo’s README? Although seems well-documented and an oversight on my part!