Dreaming Big With Google’s DeepDream Neural Net Visualization Code

Share Article

Artist Daniel Ambrosi and Engineering Collaborators To Present Large-Scale “Dreamscapes” At GPU Technology Conference

Dreamscapes by Daniel Ambrosi

Dreamscapes by Daniel Ambrosi

Artist Daniel Ambrosi will present his groundbreaking “Dreamscapes” artwork at the GPU Technology Conference, April 4-7 at the San Jose Convention Center.

Artist Daniel Ambrosi will present his groundbreaking “Dreamscapes” artwork at the GPU Technology Conference, April 4-7 at the San Jose Convention Center. Ambrosi, along with his engineering collaborators, Joseph Smarr and Chris Lamb, will discuss the workflow and process for creating these unique pieces, which are “a true collaboration between nature, man, and machine”.

The trio will hold an hour-long interactive “hangout” session at the conference on Tuesday, April 5 at 12:00. NVIDIA, the company producing the conference, has also purchased three 8 feet high by 16 feet wide backlit tension fabric structures featuring Ambrosi’s Dreamscapes to showcase at the event. The artwork -- which is the largest and highest resolution of its kind -- was created by applying a modified version of Google’s DeepDream Neural Net Visualization Code to Ambrosi’s original panoramic landscape images.

Initial attempts to utilize the open-source DeepDream software on Ambrosi’s ultra-high-resolution images were unsuccessful as the program quickly ran out of memory and crashed.

“I was extremely fortunate to be able to enlist two top-notch Silicon Valley engineers to help solve this problem: Joseph Smarr, senior staff software engineer at Google, and Chris Lamb, senior director of GPU computing software at NVIDIA,” Ambrosi said. “Joseph and Chris were enthusiastic about this challenge and kind enough to work sporadic nights and weekends on my behalf to get this code to operate successfully on my giant images.”

“Daniel has created some stunning work that’s honestly exceeded all of our expectations going in,” Smarr said. “With all the current excitement and anxiety around artificial intelligence and what it means for society, his work seems to be striking a chord. I find myself feeling an unusual sense of excitement and gratitude at having helped empower an artistic voice to be part of that conversation.”

“Daniel’s images ran on NVIDIA GPUs hosted on Amazon’s EC2 for up to 10 hours, performing as many as 90 quadrillion floating point operations per scene,” said Lamb. “It was great to experience how easy it is to stand up a compute intensive CUDA application like this, and the extreme level of detail that resulted is just thrilling to view.”

More information about the technical work required to run DeepDream on Ambrosi’s large images can be found here: http://medium.com/@jsmarr/dreaming-big-3e32ab587773

More samples of Ambrosi’s Dreamscapes can be viewed here:
http://www.danielambrosi.com/Dreamscapes

Ambrosi is available for media briefings at the GPU Technology Conference, April 4-7 at the San Jose Convention Center, or by arrangement.

Share article on social media or email:

View article via:

Pdf Print

Contact Author

Regan Daniels
Communications
+1 (650) 776-1005
Email >
Visit website