Plots.jl: Plots should check RGB values don't contain invalid values - Inexact error plotting image in gr()

Created on 21 May 2017 · 21Comments · Source: JuliaPlots/Plots.jl

gr()
x=y=linspace(-1.0,1.0,1000)
img = RGB.(x.+x',x.+x',x.+x')
plot(img)

Throw

InexactError()

 in (::Plots.##180#184)(::ColorTypes.RGB{Float64}) at /Users/solver/.julia/v0.5/Plots/src/backends/gr.jl:959
 in collect_to!(::Array{UInt32,2}, ::Base.Generator{Array{ColorTypes.RGB{Float64},2},Plots.##180#184}, ::Int64, ::Int64) at ./array.jl:340
 in _collect(::Array{ColorTypes.RGB{Float64},2}, ::Base.Generator{Array{ColorTypes.RGB{Float64},2},Plots.##180#184}, ::Base.EltypeUnknown, ::Base.HasShape) at ./array.jl:321
 in map(::Function, ::Array{ColorTypes.RGB{Float64},2}) at ./abstractarray.jl:1691
 in gr_display(::Plots.Subplot{Plots.GRBackend}, ::Measures.Length{:mm,Float64}, ::Measures.Length{:mm,Float64}, ::Array{Float64,1}) at /Users/solver/.julia/v0.5/Plots/src/backends/gr.jl:959
 in gr_display(::Plots.Plot{Plots.GRBackend}) at /Users/solver/.julia/v0.5/Plots/src/backends/gr.jl:500
 in _show(::Base.AbstractIOBuffer{Array{UInt8,1}}, ::MIME{Symbol("image/svg+xml")}, ::Plots.Plot{Plots.GRBackend}) at /Users/solver/.julia/v0.5/Plots/src/backends/gr.jl:1098
 in show(::Base.AbstractIOBuffer{Array{UInt8,1}}, ::MIME{Symbol("image/svg+xml")}, ::Plots.Plot{Plots.GRBackend}) at /Users/solver/.julia/v0.5/Plots/src/output.jl:197
 in show(::Base.AbstractIOBuffer{Array{UInt8,1}}, ::MIME{Symbol("text/html")}, ::Plots.Plot{Plots.GRBackend}) at /Users/solver/.julia/v0.5/Plots/src/output.jl:177
 in show(::Base.AbstractIOBuffer{Array{UInt8,1}}, ::String, ::Plots.Plot{Plots.GRBackend}) at ./multimedia.jl:33
 in #sprint#316(::Void, ::Function, ::Int64, ::Function, ::String, ::Vararg{Any,N}) at ./strings/io.jl:37
 in display_dict(::Plots.Plot{Plots.GRBackend}) at /Users/solver/.julia/v0.5/Plots/src/output.jl:266
 in execute_request(::ZMQ.Socket, ::IJulia.Msg) at /Users/solver/.julia/v0.5/IJulia/src/execute_request.jl:187
 in eventloop(::ZMQ.Socket) at /Users/solver/.julia/v0.5/IJulia/src/eventloop.jl:8
 in (::IJulia.##13#19)() at ./task.jl:360

GR bug

Source

dlfivefifty

Most helpful comment

it's us - fixed in https://github.com/JuliaPlots/Plots.jl/pull/1899

daschw on 22 Jan 2019

❤1 🎉1

All 21 comments

cc @jheinen

mkborregaard on 21 May 2017

Not sure this is a bug, though? The error is thrown when parsing RGB{Float64}(2.0,2.0,2.0) - but that shouldn't be defined, right? The results doesn't fit into a UInt32 which is the format used to hold the colors, and that is why the InexactError is thrown. Nor should RGB{Float64}(-2.0,-2.0,-2.0) be defined, according to the definition of the RGB type https://github.com/JuliaGraphics/ColorTypes.jl/blob/367821e5f1860ce5d10419f2c3616824c0fe2b5f/src/types.jl#L76-L79 , but half of the RGB values in the MWE are negative.

mkborregaard on 21 May 2017

Then the bug is where the error is thrown. The image itself displays without error using Images, so it's surprising that plots crashes. Probably the RGB constructor should do a check

Sent from my iPhone

On 21 May 2017, at 18:52, Michael Krabbe Borregaard notifications@github.com wrote:

Not sure this is a bug, though? The error is thrown when parsing RGB{Float64}(2.0,2.0,2.0) - but that shouldn't be defined, right? The results doesn't fit into a UInt32 which is the format used to hold the colors. Nor should RGB{Float64}(-2.0,-2.0,-2.0) be defined, according to the definition of the RGB type https://github.com/JuliaGraphics/ColorTypes.jl/blob/367821e5f1860ce5d10419f2c3616824c0fe2b5f/src/types.jl#L76-L79

—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub, or mute the thread.

dlfivefifty on 21 May 2017

Yes possibly. Do you want to open an issue on ColorTypes or should I? Closing the issue here.

mkborregaard on 21 May 2017

After the discussion here https://github.com/JuliaGraphics/ColorTypes.jl/issues/85 it transpires that downstream packages should check for RGB validity, as that cannot be done automatically. The suggestion is to use convert(RGB{N0f8}, c) where N0f8 is FixedPointNumbers.Normed{UInt8,8} - this throws an appropriate error message. This should possibly be done everywhere an RGB is used (also in PlotUtils) rather than just in gr.jl so the design will require a little thought.

mkborregaard on 21 May 2017

There is a similar issue with grayscale (or "scalar") images that might be easier to resolve. You get an InexactError if there are values outside [0,1]. Other image showing methods, e.g. the show method for image/png clamp the values. (I'd actually prefer they rescale the values and not just clamp them, but that's a separate debate :) )

FWIW, it doesn't matter to me what happens to "invalid" colors used in other contexts; e.g. it's fine to get a random error if I try to use an RGB value with negative components as the color of a line. Visualizing arbitrary arrays of floats as grayscale images is a very common and important use case though.

JeffBezanson on 20 Jan 2019

So you mean plot(Gray.(mymatrix))?
That should be easy enough to fix here https://github.com/JuliaPlots/Plots.jl/blob/57da253fa81933fc5e9f4838b0278ffc4f95de67/src/series.jl#L327-L340

mkborregaard on 20 Jan 2019

or do you mean heatmap(mymatrix, color = :Grays, clims = (0,1))

mkborregaard on 20 Jan 2019

I mean plot(Gray.(mymatrix)). See also https://github.com/JuliaImages/ImageShow.jl/issues/7.

JeffBezanson on 21 Jan 2019

Sure I'll implement the common format that's agreed on there, though I've got a comment.

mkborregaard on 21 Jan 2019

👍1

When the backend supports images, true for GR and PyPlot, Plots is not responsible for showing the images, so the only possibility we have this side is to copy the image into a new, clamped, image before passing to the backend. This is costly of course - @jheinen do you know if GR has clamping functionality we could tap into directly in the backend? And who knows about the PyPlot backend - @daschw?

For backends that don't support images, Plotly(JS), PGFPlots, InspectDR, UnicodePlots, we already copy the input and so the fix is unproblematic.

mkborregaard on 22 Jan 2019

Apparently, pyplot does the clamping, but it does not use the correct color gradient:

julia> mat = 10*rand(10, 10) .- 4.5
10×10 Array{Float64,2}:
  3.60483   -3.12377     3.33383   -0.685487  -4.21144    0.317257   1.58456   -0.59569   -0.48613    4.89837
  3.58926    0.639327    1.80893   -2.40848   -1.29764    2.01744    4.31696   -3.11538   -1.40745   -3.26707
  4.40075   -2.14372    -0.177826   0.252664   0.798815   3.42435   -1.07678    4.71595    1.83117   -4.0745
  0.999101   0.692926    0.918787   3.28246    0.248259  -0.332062  -3.56485    3.96771    4.53913   -2.82667
 -1.37461    2.02262     4.27795   -1.86656    1.91002   -4.31299    0.203546   1.31932   -1.77117   -2.14512
 -2.56787   -1.27978     1.58763   -1.54383   -0.790331  -3.3783     3.53132    3.58427    4.07261    1.88834
 -0.943326   0.610132   -3.56748    3.61886   -1.29139    1.18342    2.62486   -4.09862   -1.65401   -1.17085
  3.28029   -4.43863     1.82443   -3.74372    1.95606   -4.3604     4.39168    4.25667    4.55647   -4.47998
  4.85202    4.91935     2.82812   -0.416751   0.851981  -2.48824    4.25222    3.39537   -0.961201  -2.65786
 -2.40401   -0.0256798   2.7577     3.04841    2.71774   -1.42594   -1.23693   -0.811501   0.729003  -1.84051

julia> plot(Gray.(mat))

pyplot_image

daschw on 22 Jan 2019

ugh, is that us or native pyplot as well?

mkborregaard on 22 Jan 2019

it's us - fixed in https://github.com/JuliaPlots/Plots.jl/pull/1899

daschw on 22 Jan 2019

❤1 🎉1

Sweet - is it automatic in PyPlot then and we only need to worry about GR?

mkborregaard on 22 Jan 2019

Yes, we set the clamping values for matplotlib's imshow in the pyplot backend code.

daschw on 22 Jan 2019

OK, so your PR did fix the colors but pyplot's image still acts very weirdly on my Mac. See https://www.dropbox.com/s/zsrojfqwlve3jfn/Recording%20%2314.mp4?dl=0 the image breaks down over a certain size. Also, the clamp code doesn't seem to really be working, as there is an overflow warning from Python (see the video)

Here's a wip pr for the clamping, so far just the heatmap-approach: https://github.com/JuliaPlots/Plots.jl/pull/1900

mkborregaard on 22 Jan 2019

I cannot reproduce the weird pyplot behaviour from your video on windows.

daschw on 23 Jan 2019

Hilarious

mkborregaard on 23 Jan 2019

I'm going to implement the copy for now, but only do it if any of the values are outside (0,1). That seems like an easy and reasonable approach.

mkborregaard on 23 Jan 2019

@JeffBezanson implemented as requested in https://github.com/JuliaPlots/Plots.jl/pull/1900

mkborregaard on 23 Jan 2019

Was this page helpful?

0 / 5 - 0 ratings