Dominic Roberts

Hello! I am a Senior Applied Scientist at Microsoft, where I use machine learning to help build the best audio experience in Microsoft Teams and Azure Communication Services calls!

I previously contributed to Amazon's Just Walk Out technology and steg.ai's deep learning algorithms for invisibly watermarking images and videos.

I obtained my PhD from UIUC where I worked with Mani Golparvar-Fard and David Forsyth on construction resource activity recognition and generative modelling of 3D part hierarchies.

Email  /  Google Scholar

An augmented version of StructureNet that can re-generate parts situated at arbitrary positions in the hierarchies of its outputs.

A vision-based activity analysis method that leverages the 2D pose estimation outputs used in many state-of-the-art construction worker ergonomics analysis methods, resulting in improved performance.

A framework performing object detection, object tracking and action segmentation to automatically localize and identify earthmoving equipment and the activity they are performing in each video frame.

A prototype of an annotation tool allowing users to annotate real-world images of construction equipment with semantic segmentation masks and keypoints, given a 3D virtual model of the equipment.

A 2D human pose annotation tool adapted from CVAT that can also annotate per-frame activity labels.

Credit to Jon Barron and Unnat Jain.