Garden


The project started by assembling a dataset of over 10k artworks from open collections through their API, such as the MetMusem and the Rijksmuseum. After training a Machine Learning Model to recognize flowers of all types from all visual traditions (western painting, middle eastern carpet-making or east asian ink-wash painting, etc) they were cropped and fed to a GAN to synthetize them in new imagery. To reconvert them back to analog media, drawings and mixed media interventions were made based on the resulting images.

Generated Image
Generated Image
Generated Image
Generated Image
Generated Images, Synthetic Flowers
1024 x 1024 px
2020

Video visualizing the tSNE distribution of the garden dataset

The Mahcine Learning model for image recognition would also pick and crop details from within the artworks that resembled certain atributes of flowers, but which clearly are not. This mistake results in the automation of selecting serendipitous details from a pool of thousands of images, and from where to depart onto a new creative work.

Generated Image
Generated Image
Generated Image
Generated Image
Generated Image
Generated Image
Generated Image
Generated Image
Generated Image
Generated Image
Generated Images mistaken for flowers