GANschach: an exploration of human and machine visual perception

SkillsArtistic Research, Machine Learning

An exploration of human and machine visual perception and pattern recognition through AttnGAN-generated images. It’s called GANschach – a play on Rotschach – as it brings underlying and unconscious patterns to the surface.

As part of the Hackers & Designers Summer Academy 2019 we got a workshop titled “Bob Ross Lives” by Lenka Hamosova & Pavol Rusnak exploring GANs.

I did an experiment looking into human and machine visual perception. I ended up calling GANschach, a play on Rotschach – you know that weird ink blob test.

The process is as follows:
1) Pick random images from my phone gallery
2) Describe them in 5 – 15 words
3) Generate images from text with AttnGAN model in Runway ML
4) Feed image into Google Vision API and look at objects and labels
5) Ask people what they see in the image
6) Observe perceptions and patterns

It was exposed at the Work in Progress expo and produced some surprisingly interesting results around subjectivity in pattern recognition and threats of widespread algorithms.

See the slides used in the presentation (including what the pictures were derived from) below.