Dheera Venkatraman's ePaper Picture Frame Is a Window Into an Infinitely-Scrolling Landscape

Robotics and machine learning engineer Dheera Venkatraman has put together an electronic picture frame with a difference: it generates infinite landscape art, never-repeating, on an ePaper display.

"[EPaper] fascinates me a lot," Venkatraman explains. "It's excellent in bright sunlight, and even though most [ePaper] displays on the market are greyscale, it's extremely pleasing to the eye, and much better suited for something hanging on a wall. And so I built a new frame using [ePaper]."

The resulting digital artwork system houses a large-scale Waveshare 10.3" ePaper Display HAT in an attractive wall-mountable frame, offering a 1,872×1,404 resolution with 16 shades of grey. The display is connected to a Raspberry Pi Zero single-board computer, hidden in the rear of the frame.

Venkatraman's creation is, so far, a pretty standard display piece — but rather than download images for display, it generates them itself. "I wanted to focus on generative art," Venkatraman explains. "The first display I built is based on {Shan, Shui}* by Lingdong Huang, which procedurally generates Chinese landscape scrolls, but that are infinitely long, driven by pseudorandom numbers. The idea is the landscape on this frame is ever-changing, never repeats, and scrolls very slowly."

Huang's code, which was inspired by traditional Chinese scroll art, creates vector-format landscapes with mountains and trees based on mathematical functions, each unique and never-repeating. These vectors are then converted by Venkatraman's code for display on the ePaper panel — providing artwork that combines tradition with technology.

"For my next frame I want to build something based on [a] PyTorch implementation of DALL-E," Venkatraman adds — referring to the popular generative adversarial network, which can turn text prompts into high-quality visuals. "I'm thinking a picture frame that you can yell something at it, and it draws whatever you yell at it. We'll see how that goes, considering the limited training data available."

More details on the project are available on Venkatraman's website; source code has not been published, but Huang's {Shan, Shui}* is available on GitHub under the permissive MIT license — or can be used in-browser on Glitch.me.