Filter Feedback: Automated QA

Finding interesting configurations for a generative artwork through data analysis

Finding configurations for Filter Feedback that produce interesting results is time-consuming because generally only a tiny fraction of all randomly generated variants ends up being visually pleasing. Common problems are:

  • Configurations are too unstable; they descend into infinity (resulting in a completely black or white screen) too fast
  • Results contain strobe-like flickering, oscillating between inverted states
  • There is too little movement overall
Initial data analysis in JupyterLab

Initial data analysis in JupyterLab

Command line output of the finished tool

Command line output of the finished tool

This tool continuously generates random configurations, renders videos of them, and then judges them by analysing the frames of the video with respect to the problems outlined above. If one or more problems are detected, the configuration gets rejected.

Any configuration that was not automatically rejected that way is then presented to the user for further inspection of the video and additional data.

Art
Generative animations based on continuous application of a filter kernel
All experiments
Applied
Filter Feedback: Automated QA
Finding interesting configurations for a generative artwork through data analysis
Experiment
Increasing the chances of finding mushrooms through geospatial data analysis
Experiment
Simulating blobs of fluids with particles
Experiment
Modified JPEG encoder for generating glitchy image effects
Misc
Visual pattern derived from the 64 JPEG base blocks
Experiment
Force-based simulation of bands of particles
Experiment
Mapping line drawings onto street networks
Experiment
GPS trace replay tool
Misc
Proof-of-concept for an alternative, more powerful Are.na client
Misc
Browser extension that collects texts of how designers describe themselves
Applied
Visualization showing how a text document got written and edited over time
Experiment
Experiments with movie image data