Computer Vision Art

Helena Sarin

#prettyInGAN, (2018)

Website

This diptych is one in a series of images generated by the GAN (Generative Adversarial Network) models trained on my drawings and photographs.The first model is trained to remix my flower and portrait sketches. Subsequent models are tuned on the remixed images to achieve a texture that conveys a sense of fragility and imperfect beauty.

In my explorations of neural nets as a medium for mark making i decided to use my own datasets, of my drawings and photographs, realizing it could be quite challenging due to a smaller size of these datasets. Experimenting with GANs i found that, with some tweaks, machine learning frameworks for unpaired image-to-image translation such as CycleGAN allow me to train models on the datasets of a modest size and also generate images at a higher resolution. After creating a set of models that way, i continued to fine-tune them further - i wanted my generative artwork to be not only interesting and aesthetically pleasing but to reflect the characteristics of my analog art - improvised, bold and deeply personal…

More images from the series

part of the GAN series part of the GAN series part of the GAN series part of the GAN series part of the GAN series part of the GAN series