The Face of NVIDIA

Portrait Series Celebrating Milestone of 2,000,000 NVIDIA Developers.

Pindar Van Arman’s robot artonomous will be commemorating this milestone by training a GAN on the faces of people from the NVIDIA community, then painting a face imagined by the GAN.

1: artonomous will begin series by painting Portrait Studies of multiple individuals in the NVIDIA community in the days leading up to the event (10-20 individuals)

2: artonomous will then study the portraits both
a) at the pixel level with as many portrait photos as possible (NVIDIA’s Progressive GANs)
b) at the brush stroke level by using data gleaned in the portrait studies (Custom GAN)

3: artonomous will then use these GANs and a Style Transfer CNN to paint three AI-magined Portraits.

Important Details:

1) A live feed of the robot painting the AI-magined Portraits will be broadcast for the entirety of the event. It is estimated that 3-5 will be completed in this time frame.

2) 10-20 Painting Studies of individuals in the NVIDIA community will be completed ahead of the event.

3) As part of the exhibition, Pindar will provide a detailed work in progress of the Portrait Studies and AI-magined Portraits to describe the robot’s creative process.

Options and Miscellaneous:

1) artonomous takes about 8 hours to complete each portrait.

2) It is possible to allow portrait uploads for inclusion in the GAN before and during the event.

3) To use NVIDIA’s Progressive GAN, we will require a decent number of portraits. If for whatever reason we are unable to collect enough portraits to train the GAN, the fallback will be to use a custom GAN I wrote, combined with a Style Transfer CNN.