Error with reinterpret when setting a scale

[rsrock]

I have an image where I can display it just fine, but when I attempt to assign a scale (ScaleAutoMinMax(Ufixed16)), I run into problems. The error seems to be in map.jl:339, through take() at line 347, and up into ScaleMinMax around line 161, but I lose track of things at that point. Here's a working example, followed by the scaling and the error:

julia> using Images, ImageView

julia> img = imread("BZ363A\ 1\ redgreen.tif")
Gray Image with:
  data: 512x512x583 Array{Gray{UfixedBase{Uint16,16}},3}
  properties:
    timedim: 3
    imagedescription: <suppressed>
    spatialorder:  x y
    pixelspacing:  1 1

julia> view(img)

...works...

julia> using FixedPointNumbers

julia> img["scalei"] = ScaleAutoMinMax(Ufixed16)
ScaleAutoMinMax{UfixedBase{Uint16,16}}()

julia> view(img)

ERROR: reinterpret: expected bits type as first argument
 in map! at /Users/rrock/.julia/v0.3/Images/src/map.jl:339
 in uint32color! at /Users/rrock/.julia/v0.3/Images/src/map.jl:481
 in anonymous at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:317
 in rerender at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:689
 in view at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:327

julia>

This also happens when I take a slice of this image stack, and apply the scaling to the slice.
Any ideas where to go from here?

[timholy]

The argument you supply to ScaleAutoMinMax refers to the type you want to convert to, not the type you convert from. Try using Color; img["scalei"] = ScaleAutoMinMax(RGB24). (RGB24 is effectively the type that Cairo uses for display; it's equivalent to a Uint32, but far clearer in intent.)

[rsrock]

Ah, thanks for that tip. I was always thinking of the Scale types and map methods as "convert the specified type for display purposes". After rereading the docs, I see that it's for converting to just about anything.

However, I still seem to have the same problem here. I get an error with ScaleAutoMinMax still, while ScaleMinMax works for me. I put in a couple of @show statements in map! on line 336 here.

julia> using Images, ImageView, Color, FixedPointNumbers

julia> img = imread("BZ363A\ 1\ redgreen.tif")
Gray Image with:
  data: 512x512x583 Array{Gray{UfixedBase{Uint16,16}},3}
  properties:
    timedim: 3
    imagedescription: <suppressed>
    spatialorder:  x y
    pixelspacing:  1 1

julia> img["scalei"] = ScaleMinMax(RGB24, 0.0, 0.05)
ScaleMinMax{RGB24,Float64,Float32}(0.0,0.05,20.0f0)

julia> view(img)

...works...

mapi => ScaleMinMax{RGB24,Float64,Float32}(0.0,0.05,20.0f0)
img => Gray Image with:
  data: 512x512 SubArray{Gray{UfixedBase{Uint16,16}},2,Array{Gray{UfixedBase{Uint16,16}},3},(UnitRange{Int64},UnitRange{Int64},Int64)}
  properties:
    timedim: 0
    scalei: ScaleMinMax{RGB24,Float64,Float32}(0.0,0.05,20.0f0)
    imagedescription: <suppressed>
    spatialorder:  x y
    pixelspacing:  1 1
(ImageCanvas,ImageSlice2d: zoom = BoundingBox(0.0,512.0,0.0,512.0), t = 1)

julia> img["scalei"] = ScaleAutoMinMax(RGB24)
ScaleAutoMinMax{RGB24}()

julia> view(img)

...fails...

mapi => ScaleAutoMinMax{RGB24}()
img => Gray Image with:
  data: 512x512 SubArray{Gray{UfixedBase{Uint16,16}},2,Array{Gray{UfixedBase{Uint16,16}},3},(UnitRange{Int64},UnitRange{Int64},Int64)}
  properties:
    timedim: 0
    scalei: ScaleAutoMinMax{RGB24}()
    imagedescription: <suppressed>
    spatialorder:  x y
    pixelspacing:  1 1
ERROR: reinterpret: expected bits type as first argument
 in map! at /Users/rrock/.julia/v0.3/Images/src/map.jl:338
 in uint32color! at /Users/rrock/.julia/v0.3/Images/src/map.jl:478
 in anonymous at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:317
 in rerender at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:689
 in view at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:327

[rsrock]

I guess the real trouble is that I don't know where reinterpret is being called here.

[timholy]

After rereading the docs, I see that it's for converting to just about anything.

Yes, MapInfo is really powerful, and perhaps under-appreciated (maybe, under-documented 😄). Think of it as your gateway to "lazy" computation on images; at the very least it is convert on steroids. Not only do you specify the type that you want out, but you can also specify the algorithm. I use it all the time for sophisticated things like tingeing ROIs of grayscale images each with a different color during display. You could define a MapInfo type that blurs the image, etc.

Regarding your error, that's quite puzzling. I don't have that particular file, but:

julia> using Images, Color, FixedPointNumbers, ImageView

julia> img = grayim(rand(Uint16, 5,5,5))
Gray Image with:
  data: 5x5x5 Array{UfixedBase{Uint16,16},3}
  properties:
    colorspace: Gray
    spatialorder:  x y z

julia> map(ScaleAutoMinMax(RGB24), img)
RGB24 Image with:
  data: 5x5x5 Array{RGB24,3}
  properties:
    colorspace: Gray
    spatialorder:  x y z

julia> img["scalei"] = ScaleAutoMinMax(RGB24)
ScaleAutoMinMax{RGB24}()

julia> view(img)
(ImageCanvas,ImageSlice2d: zoom = BoundingBox(0.0,5.0,0.0,5.0), z = 1)

works for me.

Try showing mi after the take operation. Theoretically, you shouldn't even need ImageView to test this.

[rsrock]

Your test worked for me. Now I think I have a way for you to reproduce this. It does seem to need ImageView, because the final map call here works. I think it also needs an image of type Gray{Ufixed16} rather than just Ufixed16. That matches my problematic image. It's not a 3D issue, a 2D image will trigger it:

julia> using Images, Color, FixedPointNumbers, ImageView
julia> img = Image(reinterpret(Gray{Ufixed16}, rand(Uint16, 5, 5)))
Gray Image with:
  data: 5x5 Array{Gray{UfixedBase{Uint16,16}},2}
  properties:

julia> img["spatialorder"] = ["x", "y"]
2-element Array{ASCIIString,1}:
 "x"
 "y"

julia> img["scalei"] = ScaleAutoMinMax(RGB24)
ScaleAutoMinMax{RGB24}()

julia> view(img)
ERROR: reinterpret: expected bits type as first argument
 in map! at /Users/rrock/.julia/v0.3/Images/src/map.jl:336
 in uint32color! at /Users/rrock/.julia/v0.3/Images/src/map.jl:477
 in anonymous at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:317
 in rerender at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:689
 in view at /Users/rrock/.julia/v0.3/ImageView/src/display.jl:327

julia> map(ScaleAutoMinMax(RGB24), img)
mi => ScaleMinMax{RGB24,Gray{UfixedBase{Uint16,16}},Float32}(Gray{Ufixed16}(0.1293),Gray{Ufixed16}(0.96741),1.1931725f0)
RGB24 Image with:
  data: 5x5 Array{RGB24,2}
  properties:
    scalei: ScaleAutoMinMax{RGB24}()
    spatialorder:  x y

[rsrock]

P.S. I'll have to play around with those other uses of MapInfo here. Very cool.

[timholy]

Sorry this took so long. It was drattedly subtle. The call chain turned out to be map! -> take -> ScaleMinMax -> minfinite -> reinterpret. Absolutely none of those showed up in the backtrace. Inlining definitely makes debugging harder sometimes. This was enough to make me think we really need a --noinline startup mode.

ImageView creates a SubArray (to handle zoom regions), which was the final ingredient to the puzzle.

My main trick was to insert @show @which func(args...) right before the offending function calls, and then insert the same command in whichever function was being called, and so on down the chain.

[rsrock]

Thanks! Works great again. I did make it all the way to minfinite, but some testing at the REPL had fooled me into thinking that I was on a wild goose chase. This one was a pain.

I also remember seeing that the @which and @show macros were triggering different errors for me, and so I was a bit suspicious of what they were telling me. Do you ever see that?

[timholy]

Glad it's working. You may have also noted that this inspired JuliaLang/julia#9354

I haven't seen errors from @show or @which except (1) from typos or (2) too early in julia's bootstrap process.