Machine
Timing

Showcase

RE•WORK Deep Learning Summit,
Jan 2020

NYC Media Lab Summit, Sep 2019

Tools

Digital Camera, StyleGAN

Role

Developer, Photographer

Brief

Can we break the dawn of NYC with a sunrise from the Forbidden City, or can we resurrect Notre-Dame with our hundreds-of-years collective memories? “Machine Timing” is an experiment that explores the aesthetics, approaches, and constraints of using deep generative networks as a tool of time-manipulation in video footage.

Concept - A Time-based Experiment on Deep Generative Networks

Deep generative networks, such as VAEs and GANs, are often used to synthesize realistic-looking pictures. However, during experiments with such models and related techniques, I found that time-based information, along with the synthesizing subject, is also "encoded" into the trained models. As a result, if we guide the synthesizing process along the direction of a time-based factor, we can use deep generative networks to "manipulate time" in video footages.

A Sample Convertion

Demo @NYC Media Lab Summit 2019