Plots.jl: Plot an image

[yakir12]

The original idea was to plot an RGB image.
This lead to these:

using Plots, Images
img = colorview(RGB, rand(3, 100,100))

# these work:
plot(img)
heatmap(img)

x = 1:100
y = 1:100
# this draws lines, instead of changing the scale of the axis:
plot(x, y, img)
# this changes the colomap, instead of changing the scale of the axis:
heatmap(x, y, img)

[rand5]

There is a section in the documentation that shows which scales are supported in which backends. They are set using symbols. See here

[yakir12]

I assume you are referring to my comment about the scales of the axis. I'm sorry to have mislead you. I didn't mean scales as in logarithmic or linear. I meant that if I plot an image, the x and y axis will denote the pixel locations. But if I want to change that to some other specific x and y (e.g. 3:6:600) then it seems natural to me that heatmap(x, y, img) should work (or instead of heatmap use some other dedicated plotting function intended for images). But it doesn't.

[rand5]

The range of the axes is set via xlim and ylim, for example heatmap(img,xlim=(1,50)) or plot(img,xlim=(1,50)). If you want to plot every nth pixel, the most straight forward to do this is to down-sample the input data, i.e. heatmap(img[1:5:end,1:5:end]). Axes are addressed here in the docs. The behavior of heatmap(x, y, img) is the intended behavior.

[daschw]

I see what you are expecting @yakir12.
You'd like

using Plots, Images
img = colorview(RGB, rand(3, 10,10))
plot(10:10:100, 100:100:1000, img)

to produce something like this, right?:

I did this with plot(img, xformatter = x -> 10x, yformatter = y -> 100y)

That is not possible right now, because in the image recipe the x and y values are fixed to 1:m, 1:n here https://github.com/JuliaPlots/Plots.jl/blob/master/src/series.jl#L328

On the one hand, I see your point that your expected behavior would be natural (it works with heatmap without images: heatmap(10:10:100, 100:100:1000, rand(10, 10)) and I don't understand the image that is produced instead:

On the other hand, I'm not really sure what such images should look like, if the vectors x and y you want to pass are not ranges, but arbitrary vectors (like rand(10)) instead. After all, x and y denote the the discrete positions of the pixels to be colored in a matrix. So maybe using formatter or passing ticks = (locations, labels) is the cleaner solution?

[Evizero]

img = colorview(RGB, rand(3, 10,10))

FWIW you don't need to use this 2-step process to generate random images.

julia> img = rand(RGB, 10, 10)
10×10 Array{RGB{Float64},2}:
[...]

[yakir12]

You are correct @daschw. Thank you!
The x and y should only be ranges, no need for random locations (not sure what that would mean either). So your xformatter = x -> 10x solution is cool! So for an offset I'd just do xformatter = x -> 10x + x_offset.

[yakir12]

So I tried it a bit and the problem now is the number of digits after the decimal point when the formatter includes fractional numbers. So something like this:

plot(img, xformatter = x -> rand()*x, yformatter = y -> 3*rand()*y)

looks like this:

[daschw]

You could just use round in the formatter:

plot(img, xformatter = x -> round(rand()*x, 2), yformatter = y -> round(3*rand()*y, 2))

[yakir12]

🤦‍

[natschil]

@rand5: the approach you suggests seems to currently be broken if one wants to the y-axis to go from negative below to positive above, e.g. doing something like ylim=(90.0,-90.0) on a world-map to set the correct Longitudes gives me ylim scaled from 90.00 (on top) to 91.00 (on the bottom), i.e. the direction is still wrong...

[BA1437]

@natschil Could you post a minimum working example?

[natschil]

@BA1437 The image I'm currently using is a bit large, but if you have some image file the following fails for me:

using FileIO
image = load("path_to_image.tif")
plot(image, ylim=(90.0,-90.0))

The y-coordinate is always increasing when going down (I should probably add that I am using the GR backend).

[BA1437]

See https://github.com/JuliaPlots/Plots.jl/issues/1196

[daschw]

ylim always is a tuple (ymin, ymax) with ymin < ymax. If ymin >= ymax, Plots sets ymax = ymin + 1. The ymin value just appears above the ymax value for images because yflip is set true in that case.
So in your case @natschil ylim = (-90, 90) should do what you expect, right?

[daschw]

@natschil If you want the image to be flipped vertically you can do yflip = false, If you only want the tick labels to be changed from negative to positive and vice versa you can do yformatter = x -> -x

[natschil]

@daschw I don't want the image to be flipped vertically, but the coordinates used by Plots.jl (i.e. I want scatter!(xs,ys) to correctly show up on top of the image). If yflip = false is supposed to do this then there is a bug somewhere (it has absolutely no affect as far as I can tell on GR)

[daschw]

using Plots
using TestImages
img = testimage("lighthouse")

# plot the image reversing the first dimension and setting yflip = false
plot([-π, π], [-1, 1], img[end:-1:1, :], yflip = false)
# plot other data
plot!(sin, -π, π, lw = 3, color = :red)

[natschil]

This is broken in 1.0, running the code above results in the following image:

[mkborregaard]

Wow, thanks for reporting. So, the plot(img[end:-1:1,:], yflip = true) actually works in itself, it's adding the axes that makes it fail

[natschil]

Yes. Do you know what could be causing this/whether there is a workaround?

[logankilpatrick]

Hello, any update on this? It worked for me a few months ago, but now it’s not working again.

[mkborregaard]

no updates, sorry.

[logankilpatrick]

It used to work for me in Julia v0.7, without selecting a backend. Now it does not work. Any work arounds?

[mkborregaard]

Could you try the plotly backend? That used to be the default.

[daschw]

Should be fixed in #1951.