↓Skip to main content

Post-Processing for a More Organic Look

·883 words·5 mins

Linting and Typing
#

I setup Pylint and MyPy. I should’ve done this at the onset for two reasons:

  1. It’s easier to learn best practices if it’s correcting me as I go. (Fast feedback, repeatedly.)
  2. Then I wouldn’t’ve had to go through a boat load of errors and warnings in one sitting.

But it’s there now and I learned a few things in the process.

Like, Python doesn’t need you to declare that you’re accessing a variable from the global namespace unless you’re writing to it. That makes sense. If you’re reading it and there is no variable with that name in scope, it’ll check the parent scope. But if you’re writing it and there is no variable with that name in scope, maybe it means you want to create it.

I’m not going to add docstrings for everything, though. This is not that kind of project. (-:

Screenshots
#

I factored out a ScreenCapturer class so I could beef up the screenshot functionality a little bit. I added a feature that prevents overwriting images that were saved in previous sessions.

For example, if I run the insect wings app and export four images I’ll have a folder like this:

wing_000.png
wing_001.png
wing_002.png
wing_003.png

Then, if I close the app and reopen it and export one more image, wing_000.png will be overwritten. No good.

So the new ScreenCapturer class gets a list of files matching the pattern and starts after the index of the last file.

Initially, I used an algorithm which counts up. It checked the files 000, 001, 002, 003, and finally, 004. When it finds no file named wing_004.png it stores the value 4 as the index of the next screenshot. Each screenshot taken will increment the index.

So far so good.

But what if I had deleted wing_002.png and the listing now looks like this?

wing_000.png
wing_001.png
wing_003.png

Then the start index will be 2. I save one screenshot and I get wing_002.png. So far so good. But I save again and now wing_003.png is overwritten.

Unacceptable! (-:

So instead of counting up, I count down from 999. It might seem silly because it’ll count through hundreds of filenames (probably 900+ most of the time) but I care more about not overwriting the wings that I previously exported.

I guess another solution would be to determine the index to use every time a screenshot is taken. That would fill any holes left in the list of filenames and would avoid overwriting anything. That’s a totally acceptable solution as well but not what I ended up with.

Body Shapes
#

I’m running into an issue where I randomize the parameters for the base of the wing. Depending on the current values for the primary vein parameters, it can be really hard to generate a valid wing even if the values I got for the base of the wing are good.

For example, this is a reasonable shape for the wing base but the initial directions of the primary veins mean they’re almost guaranteed to overlap.

Here’s what it looks like when the initial directions are updated so as not to overlap.

And here’s an interesting-looking wing generated from that body shape:

I think the easiest solution will be to set the primary veins to perfectly horizontal lines before changing the body shape.

Actually. There’s an even easier solution. I’m doing the steps in the wrong order!

Currently I’m doing:

  1. Randomize the wing base / the body shape.
  2. Randomize the lengths of the primary veins.
  3. Randomize the curvature (including initial direction) of the primary veins.
  4. Generate the cross veins (Voronoi cells)

Step 1 passes quickly because the only check is that the resulting starting positions of the primary veins are within a certain region.

Here’s an image from an earlier post illustrating the bounding box:

So step 1 is fine.

Step 2 is where I run into problems. I’m trying to randomize the lengths of the primary veins (the number of segments) and looking for a valid wing. If the primary veins are pointed at each other then changing their lengths is not going to solve collision problems.

So I’ll just switch step 2 and step 3. I can change the lengths of the primary veins after I’ve figured out the curves of the primary veins.

Cool.

Length Constraints
#

When I’m randomizing the length of the primary veins, half the time I get a lot of primary veins with only one segment. Those results are never worth keeping.

Here are some examples:

So I’ll add constraints to automagically reject those.

…

Yeah, that helps a lot.

The above images are only varying the primary vein lengths (and regenerating cross veins).

Smoothing
#

One more thing I’d like to try. Right now the wing details look very angular and boxy. I figure I can smooth it out by blurring the image and running a threshold on it.

First attempt doing just that:

Running that image above through another blur and threshold gives me this:

So it’s the right direction. I do get a more organic look. It’s just ugly and bold right now. Maybe the first threshold should be run without the blur.

Oops. That doesn’t change anything. I’ll try with a smaller blur.

Not there yet but time to go.