trying to speed up image exports

Have been spending time running profiler to optimize code. At this point I've got things mostly trimmed down but there are a LOT of graphs, and I can spend an hour in print/saveas exporting images. I was looking at this post

https://www.mathworks.com/matlabcentral/answers/1906890-how-to-save-a-figure-to-a-png-file

and noticed that export was around 5-10% faster than using saveas. I've tried saving as svg instead of png, sometimes it is faster, sometimes its slower (for the same plots/dataset). There are some old posts from 2010-14 on speed-ups, but I've got to imagine those aren't as fast now that Matlab has native functions for this.

Is there any guidance on how to get improvements on these processes? I can be flexible on file format/resolution etc. for anything that going to be at least 10% faster.

[Walter Roberson]

To clarify:

You have individual axes that you wish to save as images? Or are they figures that you wish to save as images (including the possibility of subplot() or tiledlayout() ?)

[Justin Bell]

@Walter Roberson good point, to clarify the use case is along the lines of

X=rand(1000,1); Y = rand(1000,1);
figure()
scatter(X,Y)
saveas(gcf, outname.png, 'png')

about 1000 times inside of some loops. currently not using subplots or tiled layouts, so a robust solution is always nice - but if something works for all plots but those, that is still a win for me.

I need to do a few more tests, but interstingly enough, using saveas ppm was much slower, but when I duplicate your test using imwrite it is faster. I couldnt test this with exportgraphics(gcf, 'outfile.fileType') because export graphics doesnt support ppm

Answer 1

[Walter Roberson]

This test suggests that writing to ppm might potentially be faster, with writing to JPEG2 being notably slower.

data = randi([0 255], 128, 160, 3, 5, 'uint8');

formats = [".bmp"; ".jpg"; ".jp2"; ".pbm"; ".png"; ".ppm"; ".ras"; ".tif"];
n_formats = length(formats);
times = zeros(n_formats,1);

for J = 1 : n_formats
    form = formats(J);
    tic
    for K = 1 : size(data,4)
        imwrite(data(:,:,:,K), "test"+K+form);
    end
    times(J) = toc;
end
T = table(times, 'rownames', formats)

T = 8x1 table
             times  
            ________

    .bmp    0.028122
    .jpg    0.016113
    .jp2     0.33083
    .pbm    0.025394
    .png    0.028225
    .ppm    0.012019
    .ras    0.015207
    .tif     0.11636

[Walter Roberson]

For saveas(), jpeg is fastest, followed by tiff.

With jpeg being a lossy format, it is often not a good idea to save technical images as jpeg.

formats = [".bmp"; ".jpg"; ".pbm"; ".png"; ".ppm"; ".tif"];
n_formats = length(formats);
times = zeros(n_formats,1);


X = rand(1000,1); Y = rand(1000,1);
scatter(X,Y)
fig = gcf;

for J = 1 : n_formats
    form = formats(J);
    tic
    for K = 1 : 5
        saveas(fig, "test"+K+form);
    end
    times(J) = toc;
end

Warning: Image format 'pbm' will not be supported in a future release.

Warning: Image format 'pbm' will not be supported in a future release.

Warning: Image format 'pbm' will not be supported in a future release.

Warning: Image format 'pbm' will not be supported in a future release.

Warning: Image format 'pbm' will not be supported in a future release.

Warning: Image format 'ppm' will not be supported in a future release.

Warning: Image format 'ppm' will not be supported in a future release.

Warning: Image format 'ppm' will not be supported in a future release.

Warning: Image format 'ppm' will not be supported in a future release.

Warning: Image format 'ppm' will not be supported in a future release.

T = table(times, 'rownames', formats)

T = 6x1 table
             times 
            _______

    .bmp     1.5454
    .jpg    0.79257
    .pbm     1.6861
    .png     1.2041
    .ppm     3.9328
    .tif    0.83244

When I tried this on my iMac using R2024b, only [".bmp"; ".jpg"; ".png"; ".tif"] were supported and the times were significantly longer,

times

___________

.bmp 3.354304789

.jpg 3.275275412

.png 3.719371674

.tif 3.376153964

It is difficult to draw clear conclusions from that, except that jpeg might be fastest and png might be slowest of the four.

[Justin Bell]

Thanks for the detailed response and useful code snippet! I've been experimenting using this with different plot types. Interestingly enough, different plots will save out at different relative rates. For example, if there are multiple colors in the scatter or if there are alpha layers, the faster plot will be different.

The overall answer seems like "it depends on the machine and type of plot" but now using the loops you suggested, I can dial things in. Interestingly, exportgraphics tends to match saveas in terms of which formats are fastest, but imwrite (which I don't really use) seems inversely related.

[DGM]

FWIW, this is much faster for me.

formats = [".bmp"; ".jpg"; ".pbm"; ".png"; ".ppm"; ".tif"];
n_formats = length(formats);
times = zeros(n_formats,1);
X = rand(1000,1); Y = rand(1000,1);
scatter(X,Y)
fig = gcf;
for J = 1 : n_formats
    form = formats(J);
    tic
    for K = 1 : 5
        outpict = frame2im(getframe(gcf));
        imwrite(outpict, "test"+K+form);
    end
    times(J) = toc;
end

T = table(times, 'rownames', formats)

T = 6x1 table
             times 
            _______

    .bmp    0.68261
    .jpg    0.45181
    .pbm    0.53088
    .png    0.50899
    .ppm    0.45058
    .tif     0.4842

These are the times on a 12 year old computer that came out of a dumpster. PNG with saveas() doesn't look as bad, but using getframe()/imwrite() is faster

% using saveas()
%             ______
%     .bmp    2.5871
%     .jpg    12.795
%     .pbm    4.7508
%     .png    2.4091
%     .ppm    11.929
%     .tif    2.3482

% using getframe()/imwrite()
%             _______
%     .bmp    0.87231
%     .jpg    0.25393
%     .pbm     0.2676
%     .png    0.36382
%     .ppm    0.68319
%     .tif    0.21874

There seems to be a lot of variability using saveas(), but the general theme is that they're all slower.

% using saveas() (another run)
%             ______
%     .bmp    5.6396
%     .jpg    1.9731
%     .pbm    4.5735
%     .png     2.497
%     .ppm    11.676
%     .tif    7.4597

Why are they slower? We shouldn't expect that the images written by the two methods are the same size, but even accounting for the difference in size, saveas() is still a lot slower.

Answer 2

[Subhajyoti]

Hi @Justin Bell,

You can export the figures to FIG-files ('.fig') using the “savefig” function. This gives better performance at the major performance bottleneck, which is exporting the figure to external files.

Here, in the following implementation, I have clocked the running time for the function.

tic

% generate random data. Plot them and save the plot to PNG file
for i = 1:1000
    %Preparing data for plotting
    x = 1:1000;
    % generate 1000 random data following normal distribution
    y = 255*randn(1,1000);

    % Save the plot to PNG file without displaying it
    fig = figure('Visible','off');
    plot(x,y);
    title('Random data','Interpreter','latex');
    xlim([0 1000]);
    ylim([100 255]);
    xlabel('X','Interpreter','latex');
    ylabel('Y','Interpreter','latex');
    set(gca,'TickLabelInterpreter','latex');
    
    % Save the plot to PNG file
    filename = ['plot' num2str(i) '.fig'];
    savefig(fig,filename);
    close(fig);
end

toc

Elapsed time is 119.392816 seconds.

The above implementation exports at '500 figures/minutes', which is around 5x faster than PNG-format. You can further improve the output time by setting the ‘Visible’ parameter of the ‘figure’ object to ‘off’.

You can open the saved figure using ‘openfig’ from any MATLAB script or Command-Line as follows:

openfig("plot1.fig", "visible")

ans = 
  Figure (1) with properties:

      Number: 1
        Name: ''
       Color: [1 1 1]
    Position: [1629 1510 583 437]
       Units: 'pixels'

  Use GET to show all properties

Refer to the following resources to know more importing and exporting Figures in MATLAB:

• https://www.mathworks.com/help/matlab/ref/savefig.html
• https://www.mathworks.com/help/matlab/ref/openfig.html

Additionally, you can refer to the following resources to know more about saving Plot as Image or Vector Graphics File in MATLAB:

https://www.mathworks.com/help/matlab/creating_plots/saving-your-work.html