CVPR 2020 · lead author
UniPose: Unified Human Pose Estimation in Single Images and Videos
A single-stage architecture for pose estimation in images and video, without the separate detection step earlier methods relied on.
Research
Research
My Ph.D. work was on human pose estimation: single-stage neural architectures that locate the body's joints in images and video. It led to UniPose, OmniPose and Full-BAPose.
Google Scholar
500+
citations across 20+ peer-reviewed publications, and one patent in applied AI
Selected publications
IEEE TPAMI 2021 · lead author
UniPose+: A Unified Framework for 2D and 3D Human Pose Estimation in Images and Videos
The journal extension of UniPose, covering 2D and 3D pose estimation in one framework.
BMVC 2021 · lead author
OmniPose: A Multi-Scale Framework for Multi-Person Pose Estimation
Extends the approach to scenes with several people.
MDPI Sensors 2023 · lead author
Full-BAPose: Bottom-Up Framework for Full Body Pose Estimation
A bottom-up framework for full-body pose estimation, beyond a sparse set of keypoints.
Themes
Research areas
Multi-scale vision and WASP
Waterfall Atrous Spatial Pooling (WASP) is a module that lets a network use fine detail and wider context in a single pass. I helped introduce it and used it across several papers.
Single-stage pose estimation
Most strong pose estimators of the time detected people first and estimated joints second. My work predicts the joints directly.
From 2D to full body
UniPose handled 2D keypoints, UniPose+ added 3D, and Full-BAPose extended the approach to the full body.
Current interests
Since the Ph.D. my focus has moved to applied work: multi-agent systems, retrieval and evaluation.