Totally crazy behaviour of figures. Axes going around through the image!

Hello,

I'm working on a GUI in matlab and I ended up with a really strange behaviour in my code. I manage to isolate the problem in a single script, which is the following:

fig = figure('Name','Spatial modes', ...
    'numbertitle', 'off', ...
    'ToolBar', 'none', ...
    'menu','none','units','normalized',...
    'outerposition',[0 0 1 1]);
title('whatever')

for i = 1:100
    set(0,'CurrentFigure',fig)
    cla
    imagesc(rand(10))
    caxis auto
    hand.im_caxis.val = caxis;

      colormap hot
      colorbar
      axis ij equal tight
      set(gca,'color','none')
      title(num2str(i))
      drawnow
      pause(0.1)
  end

If you try to run it, you will see that the axes of the figure shift to the right every time the image is updated. After a dozen of shifts, they start shrinking until they are invisible and eventually they go back to the original size.

What is going on? What am I missing?

Answer 1

[Alessandro Masullo]

I contacted the MathWorks Support and they confirmed me that this is a bug of the R2015a version.

They also suggested me a simple workaround to avoid this problem. If you experience a similar issue, try to call the colorbar line after the axis line:

   axis ij equal tight
   colorbar

This will prevent the axes to move randomly.

Answer 2

[the cyclist]

I ran your code in R2016a, and did not have the "disappearing figure" you show. It worked fine.

[Alessandro Masullo]

I ran the code in matlab R2015a and the result is shown in the attached gif. Is there a bug? And if so, is there a workaround?

[Walter Roberson]

I tested in R2015a on OS-X and do not get a shrinking plot. The labels on the colorbar move around a bit but the sizes do not change (the change in label reflects the fact that imagesc() depends upon the range of values in the data but since your data is randomly generated, the range of values changes from random matrix to random matrix.)

Answer 3

[Image Analyst]

I also do not observe the effect in R2016a. The best workaround is to upgrade to the latest release. Another workaround might be to get the 'Position' property of the axes the first iteration, and then use that to set it in subsequent iterations. So if it got changed somehow, that should fix it. Try adding this code:

  imagesc(rand(10))
  ax = gca;
  if i == 1
    position = ax.Position;
  else
    ax.Position = position;
  end

[Alessandro Masullo]

Hello Image Analyst,

This must definitely by a bug. I tried this code:

close all force
clear
fig = figure('Name','Spatial modes', ...
    'numbertitle', 'off', ...
    'ToolBar', 'none', ...
    'menu','none','units','normalized',...
    'outerposition',[0 0 1 1]);

for i = 1:100
    imagesc(rand(10))
%     foo = caxis;
    colormap hot
    colorbar
    axis ij equal tight
    set(gca,'color','none')
    title(num2str(i))
    drawnow
    pause(0.1)
end

and the figure is updated nicely. If you uncomment the line "foo = caxis;" then the image is updated in that crazy way I showed in the image.

Unfortunately, I cannot use your solution because in my code I don't have a simple for loop, but a function that updates the figure based on a callback, therefore I don't know when it is generated for the first time.

[Image Analyst]

I can't test your version, but if it was a bug then it's fixed now since I don't observe it with R2016a.