Research

I hold a Ph.D. in Electrical & Computer Engineering from RIT. My research centers on human pose estimation, multi-scale feature aggregation, and single-stage architectures. 20+ peer-reviewed papers, 500+ citations, and a patent in applied AI.

20+

peer-reviewed publications.

500+

citations across the pose-estimation and vision literature.

patent in applied AI.

Research themes

Multi-scale vision & WASP.

I helped introduce and extend the Waterfall Atrous Spatial Pooling (WASP) module, a multi-scale context aggregation block that informs several of my papers on pose estimation. The architectural question: how do you let a network see fine-grained joints and large-scale body context in a single pass?

Single-stage human pose estimation.

Most strong pose estimators at the time relied on two-stage, detection-first pipelines. My line of work (UniPose, OmniPose, Full-BAPose) pursues unified, single-stage formulations that predict joints (and sometimes full body surface) directly.

From 2D to full-body.

A trajectory from 2D keypoints (UniPose) to 3D pose (UniPose+) to full-body / bottom-up formulations (Full-BAPose). The thread is architectural: how far can you push a single-stage, context-aware design?

Applied ML & production systems.

Post-Ph.D., my research interest has shifted toward applied multi-agent systems, retrieval-augmented coaching pipelines, and responsible-AI evaluation, the underpinnings of Athlete Space and Wizard AI.

Selected publications

UniPose Figure 1: pose estimation examples

CVPR 2020

UniPose: Unified Human Pose Estimation in Single Images and Videos

Bruno Artacho, Andreas Savakis.

A unified, single-stage architecture for human pose estimation across still images and video. Introduces a context-aware design with multi-scale feature aggregation and achieves competitive accuracy without the two-stage detect-then-estimate pipeline used by prior work.

PDF Citations

Pose estimation examples from the UniPose+ line of work

IEEE TPAMI 2021

UniPose+: A Unified Framework for 2D and 3D Human Pose Estimation in Images and Videos

Bruno Artacho, Andreas Savakis.

Journal extension of UniPose that unifies 2D and 3D pose estimation inside a single architectural framework. Published in IEEE Transactions on Pattern Analysis and Machine Intelligence, one of the field's most cited venues.

IEEE Xplore Citations

OmniPose Figure 1: multi-person pose estimation examples

BMVC 2021

OmniPose: A Multi-Scale Framework for Multi-Person Pose Estimation

Bruno Artacho, Andreas Savakis.

Extension of the WASP-based line of work to multi-person pose estimation, further pushing single-stage, multi-scale architectures for complex multi-subject scenes.

arXiv Citations

Full-BAPose Figure 1: full-body pose estimation examples

MDPI Sensors 2023

Full-BAPose: Bottom-Up Framework for Full Body Pose Estimation

Bruno Artacho, Andreas Savakis.

A bottom-up, single-stage framework for full-body pose estimation, extending the line of work beyond sparse keypoints to full-body representations suitable for real-world sensing applications.

Sensors Citations

Full record

Google Scholar profile.

Complete publication list, citation counts, and h-index metrics maintained on Google Scholar.

Open Google Scholar