Example of how to script figure creation with inkscape and python.

Generally, when creating figures, I use matplotlib and seaborn to create and save plots as svgs (an open standard for vector graphics) and combine them, possibly with some text, using svgutils. However, for one of my research projects occasionally I want to include .png images in my figures. These are raster graphics and so suddenly I have to worry about the resolution of the images included in my figure, because I want them to be visible and not alias. For smaller images, I might simply down-sample them until they have the appropriate resolution, but for large images, this won't do.

A standard way to handle this problem is to use a vector graphics editing program, such as inkscape or Adobe Illustrator, and link the images, then, when the figure is finished, export it to a pdf and set the dpi then. This works, but I try to script as much of my workflow as possible, especially while writing a paper, when I might go through many different versions of a figure while trying to find the right one. So I managed to come up with the following solution, in which:

1. The user creates a template svg file in inkscape, which defines where the different images will go (as well as any additional image content, such as text or shapes). This template has dummy paths, (IMAGE1, IMAGE2, etc) which will be replaced with the png images.
2. sed replaces those dummy paths with the path to the actual image.
3. inkscape embeds those png files in the svg with a user-specified dpi.
4. inkscape converts that svg to a pdf.

Snakemake is used to manage everything after step 1.

• Python 3, with following packages, all of which can be installed via pip:
  • Snakemake, a python-based workflow manager.
  • flatten-dict, for easily flattening nested dictionaries.
  • xmltodict, for parsing xml and svg files.
• Inkscape, an open source vector graphics editor.
• sed, "stream editor", a non-interactive command-line text editor (this is probably already installed on your system if it's Mac or Linux).

This example repo already includes the template.svg and png files to embed, curie.png and einstein.png. Our template also inlcudes an already-embedded png file (a patch of white noise), as well as text and shapes. You can open this up with inkscape to see what it looks like; you'll notice four "Linked image not found" boxes of different sizes (you can also open up the file with your browser, but at least for me, those boxes are just not rendered).

Before running anything, we need to know where your inkscape preferences file is. If you open inkscape and go to Edit -> Preferences -> System, you can see the User preferences field underneath the System info section. Copy this path and paste it INKSCAPE_PREF_FILE at the top of the Snakefile.

To embed the image, we just use snakemake: snakemake -n -prk figure_dpi-10.pdf figure.pdf. This is a dry-run (because of the -n flag), so you can see what steps will be run. Here, we're creating two versions of the pdf file: one with the png files embedded at 10 dpi, one with them embedded at their full dpi. To run, simply replace the -n with -j 1, which tells snakemake to run only one process at a time.

At the end of this, you'll have two pdf files, as well as several intermediate svgs. You can use du -hs ./*pdf to see that the dpi-10 version is much smaller, and you can see why when you open the files up to view -- 10 dpi is really low! The use of 10 dpi is just for example purposes, it's way too low to actually use for anything. Simply change the number in the path (e.g., figure_dpi-300.pdf) to change it.

All the code can be found in Snakefile, which contains two python functions and three snakemake rules. All are commented and documented, so you're encouraged to read through them, but you should only need to understand the snakemake rules:

• link_bitmap_images: use sed to replace our dummy paths ("IMAGE1", etc) with the paths to the png files.
• embed_bitmaps_into_images: find the ids of the linked images (taking advantage of the fact that svgs are just a special type of xml file), then use inkscape's scripting ability to embed them all at the specified dpi.
• convert_to_pdf: convert an svg to a pdf. This is so simple it probably doesn't need to be a rule, but this way we can script the whole process.

If you'd like to use this trick for your own work, here's a handful of things you might like to do:

• Rearrange things in the template: simply edit this file in inkscape, moving them around as you see fit. Add any amount of additional svg elements you'd like (including embedding other raster images directly); they won't interfere with this process.
• Link and embed other images using this process: copy one of the existing linked images, right click on it and select image properties. In the pane that opened up, set the URL to IMAGE3 (or whatever the next sequential image is). Then, in the Snakefile, go to the link_bitmap_images rule and add the path to the image you'd like embedded after 'curie.png' in the input: section.
  • Similarly, if you'd like to replace one of the existing images, just change the path in the input: section of that rule.
  • Make sure the aspect of your placeholder matches what you want for your embedded image. As you can see in the example, the embedded image will be stretched to match the placeholder.
• Create a new template: just open up inkscape and create a new svg with some placeholders! I do this by linking the image I want to embed, then resizing it to my desired size (starting from the image I want makes it easier to keep the aspect consistent). Then, as above, change the URL to IMAGE1.
  • I've sometimes been unable to set the URL to an invalid one like this. If that's the case for you, save your template svg and open it with a text editor. Find your image (I recommend searching by its path), and edit the xlink:href field to "IMAGE1" (note the quotes!).
• Change the output dpi: when running the snakemake rule to create the figure, simply change the number after dpi- in its path.
• Combine with plots: if you want to combine the figure created here with plots created by matplotlib or other python plotting libraries, I recommend saving those as svg files and then using svgutils to combine the two. I would insert a rule after link_bitmap_images and before embed_bitmaps_into_images which combines the two svgs before embedding the bitmaps. svgutils will not display the linked image, so to make developing this easier, I'd recommend creating a rectangle with black stroke and empty fill of the same size as your placeholder, and place it in the same location; that way, you'll know where the placeholder is. You can also use svgutils to programmatically add text, though I'm pretty sure svgutils assumes there are always 90 pixels per inch, so you'll need to take that into account when figuring out sizes.
• Note: I'd save the svg as a "Plain svg" rather than an "Inkscape svg" (in the dropdown menu, when you go to save), since this strips out some unnecessary tags from the svg file, which makes it easier to go through if you edit the svg as a text file.
• Note: When creating your template, I think its helpful to set the display units to go to File -> Document Properties and set the Display units: to px, and then, under Scale, set both the x and y values to 1.0. This makes it easier to ensure your sizes are consistent when editing your image in multiple programs (e.g., with both inkscape and the python svgutils library).