Alex J. Champandard

nips18_RegardingPablo_A[hd] (1)

Regarding Pablo (2018)

This series of renderings is inspired by and based on Picasso’s cubist work during his African Period — featuring bright colors, chaotic feel yet rhythmic texture, and quickly painted lines. The original painting is deconstructed into basic elements and stripped of its meaning, then brought back to life again as a generative texture with infinite possibilities using the latest algorithms in neural image synthesis.  It’s intended as a study to understand, capture, and reproduce the essence of Picasso’s painting style in one specific piece, La Grande Odalisque d’après Ingres (1907), which for the first time is demonstrated to be possible with techniques based on deep learning.

While neural style transfer techniques captured the imagination of the general public and machine learning researchers, they have yet to find their way into regular workflows of professional artists.  Part of the challenge is capturing sharp edges of pencil or other sketches, or in fact any kind of highly context-sensitive patterns in the image.  This series started as an exploration of the biggest challenges of the field, in particular this Picasso style, and in the process discovering what combination of ideas (new or existing) were necessary to make this happen.

The script used to produce these artworks is based on an open source project that’s available with documentation at this URL: https://github.com/alexjc/neural-imagen

nips18_RegardingPablo_B[hd] (1)