Augmented Reality / Machine Learning / Virtual Reality / Tools

Volume

Stack

Python, Javascript, GLSL, C#, HLSL
Three.js, Unity3D, Unreal Engine, Blender, Tensorflow
Heroku, Firebase

About

Volume is a tool for reconstructing a single 2D image or video in 3D space.

Using state-of-the-art machine learning research, Volume is able to generate a 3D asset from a single view.

Volume was my graduate school thesis in ITP at NYU, which I developed with Shirin Anlen.

Machine Learning: Based on state-of-the-art machine learning research, Volume is included a fine tuned monocular depth estimation model that is used to reconstruct a single 2D image or video in 3D space.

API/volume.gl: Volume was built as an end-to-end solution allowing anyone to easily generate a 3D asset and use it in 3D environments using either our inference API or make.volume web app.

Thesis Presentation

Experiments

In the process of developing Volume, I built a number of experiments to explore the capabilities of using depth estimation to reconstruct archival footage.

JFK speech

Inside Pulp Fiction

Inside Pulp Fiction is an experiment that uses Volume to reconstruct Pulp Fiction's iconic dance scene in Augmented Reality.

Github Repository

ReTouch

ReTouch is an OpenGL application that enables editing and retouching of images using depth-maps in 2.5D.

ReTouch

Github Repository

Press

Credits

Developed with ~shirin anlen

Additional Links