GAN Weaver (2019) is part of the ongoing #latentDoodles project. For the last few months I’ve been training SNGAN-projection on various datasets of my sketches, doodles and random (personal) datasets which i vary over the time to inject the cross contamination and bias. The separate tiles are generated to be assembled randomly into a single piece where i control the dimensions, checkpoints/classes to use, dis/similarity and then curate the results.
Though purely generative, the idea behind the series is to create organic marks evocative of human hand, but with hints of generative computational origin through pattern repetition and grid. Vis-à-vis abstract nature of generated images, i work carefully on image captioning, to give them additional (verbal) dimension, this particular image is like a tapestry produced by a loom.
