Handtrack.js: Building gesture-based interactions in the browser using TensorFlow.js

Victor Dibia (Cloudera Fast Forward Labs)

11:50am–12:30pm Thursday, October 31, 2019

Location: Grand Ballroom E

JavaScript

Who is this presentation for?

Frontend web (JavaScript) developers, user experience designers, and data scientists

Level

Intermediate

Description

While JavaScript continues to be the most-used programming language, until recently, there’s been limited frameworks for machine learning that cater to this audience. With the advent of TensorFlow.js—a library for developing and training ML models in JavaScript for deployment in browser or on Node.js—things are changing. When running inside the browser, TensorFlow.js uses the GPU of the device via WebGL to enable fast parallelized floating point computation. In Node.js, TensorFlow binds to the TensorFlow C library, enabling full access to TensorFlow.

In the browser, ML can enable truly novel forms of interactions while reaping the benefits associated with on-device computation such as reduced latency for interactive applications, reduced model distribution costs, and enhanced privacy, as data is no longer sent to remote servers for analysis.

Victor Dibia provides an overview of the TensorFlow.js library, benchmarks performance results for image tasks in the browser, and provides a live demonstration of Handtrack.js—a library for prototyping real-time hand-tracking interactions. Handtrack.js is powered by an object detection neural network (MobileNetV2, SSD) and allows you to predict the location (bounding box) of human hands in an image, video, or canvas HTML tag. You’ll learn the steps and best practices for deploying a neural network model in the browser from data collection, model training, and model conversion to TensorFlow.js WebModel format, model hosting, and inference.

What you'll learn

Understand the state for machine learning in the browser using TensorFlow.js 1.0
Learn the steps and best practices for deploying a neural network model in the browser from data collection, model training, and model conversion to TensorFlow.js WebModel format, model hosting, and inference
See examples of hand-tracking interactions prototyped in the browser to help understand UX concerns and future possibilities

Victor Dibia

Cloudera Fast Forward Labs

Victor Dibia is a research engineer at Cloudera’s Fast Forward Labs, where his work focuses on prototyping state-of-the-art machine learning algorithms and advising clients. He’s passionate about community work and serves as a Google Developer Expert in machine learning. Previously, he was a research staff member at the IBM TJ Watson Research Center. His research interests are at the intersection of human-computer interaction, computational social science, and applied AI. He’s a senior member of IEEE and has published research papers at conferences such as AAAI Conference on Artificial Intelligence and ACM Conference on Human Factors in Computing Systems. His work has been featured in outlets such as the Wall Street Journal and VentureBeat. He holds an MS from Carnegie Mellon University and a PhD from City University of Hong Kong.