Image Decompiler

August 27, 2018 · View on GitHub

Arranges triangles in order to minimize the difference between the framebuffer and the source image.

Currently only handles PNG images.

How It Works:

The goal of the image decompiler is to produce:

The process of creating the list of triangles is as follows:

Read in parameters from user, such as source png image and maximum triangle count
Create N primary and N secondary canvases
Create N empty triangle lists
Add a small triangle with random color and random position to the current most optimal list
Set the frame counter to 0
Create N - 1 variations of the current most optimal list. These variations are each a random 'distance' and random 'direction' away parameter-wise, and biased towards changing the most recently added triangle
For the N primary canvases: draw the triangle lists into the canvases
For N secondary canvases: draw the absolute value of the difference between the source image and the primary canvas into the secondary canvas
For N secondary canvases: determine the average value of the difference (top mipmap value)
Choose the canvas with the lowest average difference and set it as the current most optimal list
Save a screenshot once every M minutes
Handle user input
Add one to the frame counter
If the triangle list is full or the frame counter is less than F: goto 4
goto 3