Overlay doesn't work with PyPlot

[bjarthur]

i can ImageView.view an Overlay just fine, but i cannot PyPlot.imshow. is it because raw and separate don't work?

julia> using Images, ColorTypes, TestImages

julia> mandril = testimage("mandril_gray")
Gray Images.Image with:
  data: 512x512 Array{ColorTypes.Gray{FixedPointNumbers.UFixed{UInt8,8}},2}
  properties:
    imagedescription: <suppressed>
    spatialorder:  x y
    pixelspacing:  1 1

julia> ovr = Overlay((mandril,mandril),(RGB(1,0,0),RGB(0,1,0)));

julia> using PyPlot

julia> imshow(ovr)
2016-03-25 08:53:03.909 julia[5481:41933] setCanCycle: is deprecated.  Please use setCollectionBehavior instead
2016-03-25 08:53:03.918 julia[5481:41933] setCanCycle: is deprecated.  Please use setCollectionBehavior instead
ERROR: PyError (:PyObject_Call) <type 'exceptions.TypeError'>
TypeError(u'Image data can not convert to float',)
  File "/Users/arthurb/.julia/v0.4/Conda/deps/usr/lib/python2.7/site-packages/matplotlib/pyplot.py", line 3022, in imshow
    **kwargs)
  File "/Users/arthurb/.julia/v0.4/Conda/deps/usr/lib/python2.7/site-packages/matplotlib/__init__.py", line 1811, in inner
    return func(ax, *args, **kwargs)
  File "/Users/arthurb/.julia/v0.4/Conda/deps/usr/lib/python2.7/site-packages/matplotlib/axes/_axes.py", line 4943, in imshow
    im.set_data(X)
  File "/Users/arthurb/.julia/v0.4/Conda/deps/usr/lib/python2.7/site-packages/matplotlib/image.py", line 449, in set_data
    raise TypeError("Image data can not convert to float")

 [inlined code] from /Users/arthurb/.julia/v0.4/PyCall/src/exception.jl:81
 in pycall at /Users/arthurb/.julia/v0.4/PyCall/src/PyCall.jl:344
 in imshow at /Users/arthurb/.julia/v0.4/PyPlot/src/PyPlot.jl:460

julia> raw(ovr)
ERROR: auto_unbox: unable to determine argument type
 in raw at /Users/arthurb/.julia/v0.4/Images/src/core.jl:318

julia> separate(ovr)
ERROR: MethodError: `reinterpret` has no method matching reinterpret(::Type{FixedPointNumbers.UFixed{UInt8,8}}, ::Images.Overlay{FixedPointNumbers.UFixed{UInt8,8},2,2,Tuple{Images.Image{ColorTypes.Gray{FixedPointNumbers.UFixed{UInt8,8}},2,Array{ColorTypes.Gray{FixedPointNumbers.UFixed{UInt8,8}},2}},Images.Image{ColorTypes.Gray{FixedPointNumbers.UFixed{UInt8,8}},2,Array{ColorTypes.Gray{FixedPointNumbers.UFixed{UInt8,8}},2}}},Tuple{Images.ScaleMinMax{Float32,FixedPointNumbers.UFixed{UInt8,8},Float32},Images.ScaleMinMax{Float32,FixedPointNumbers.UFixed{UInt8,8},Float32}}}, ::Tuple{Int64,Int64,Int64})
Closest candidates are:
  reinterpret{T,S,N}(::Type{T}, ::Array{S,N}, ::NTuple{N,Int64})
  reinterpret{T,Tv,Ti,N}(::Type{T}, ::SparseMatrixCSC{Tv,Ti}, ::NTuple{N,Int64})
  reinterpret{T,S}(::Type{T}, ::S)
  ...
 in separate at /Users/arthurb/.julia/v0.4/Images/src/core.jl:408

[mronian]

I get the reinterpret has no method matching error sometimes while using the save method on the images returned from the histeq method in #458.

[timholy]

Ah, overlay images shouldn't even be using reinterpret---overlays are "lazy overlays," meaning they don't require you to combine the arrays. (Intended to make it possible to construct overlays with TB-sized images, and have them display on-the-fly.) The downside, of course, is that the more fancy types one has, the greater the risk that some operations haven't been defined.

Probably the easiest approach would be to create a specialized method of separate for overlay types, implemented as separate(copy(img)) or something.

As for the raw error, I haven't looked at that. Which line does 318 correspond to?

[bjarthur]

thanks tim! and not just for your work on the Images and Colors family of packages, but your contributions to julia Base in general. i am continually perplexed as to how you manage to run a lab as well!

as a note, Overlays now do work with PyPlot, but in the specific example above, one must explicitly convert to Float, and make sure that the color dimension is last:

imshow(convert(Array{Float32,3}, separate(ovr)))