Hueihan Jhuang

Also known as Huei-han Jhuang

Hueihan Jhuang is a computer vision researcher who co-created HMDB, a 2011 video database of about 7,000 annotated clips in 51 human action categories. Earlier work included a 2007 ICCV paper with Thomas Serre, Lior Wolf and Tomaso Poggio that built an action recognition system on a model of motion processing in the visual cortex, and a 2010 Nature Communications paper from MIT's McGovern Institute on a trainable vision system that scored mouse behavior in home cages as accurately as human observers. An MIT PhD followed in 2011, with the dissertation "Dorsal Stream: From Algorithm To Neuroscience" advised by Poggio. At the Max Planck Institute for Intelligent Systems, Jhuang was first author of the 2013 J-HMDB study, which added human joint annotations to HMDB and found that high-level pose features greatly outperform low- and mid-level features for recognizing actions. With the rest of the HMDB51 team and the UCF101 dataset team, Jhuang shared the Everingham Prize at ECCV 2022.

Timeline

2007

A biologically inspired action recognizer

With Thomas Serre, Lior Wolf and Tomaso Poggio, Jhuang publishes "A Biologically Inspired System for Action Recognition" at ICCV 2007. The system extends a neurobiological model of motion processing in the visual cortex into a hierarchy of spatio-temporal feature detectors, and it achieves the highest results reported to date on public action datasets.
September 2010

Automated mouse behavior analysis

In September 2010, Nature Communications publishes Jhuang's first-author paper from the Department of Brain and Cognitive Sciences and McGovern Institute at MIT on a trainable computer vision system that analyzes mouse behavior in home cages. The system scores behavior as accurately as human observers, and the team releases the software and its annotated video database for free.
2011

A PhD under Tomaso Poggio

In 2011, Jhuang completes a PhD at the Massachusetts Institute of Technology with the dissertation "Dorsal Stream: From Algorithm To Neuroscience," advised by Tomaso A. Poggio.
2011

The HMDB video database

At ICCV 2011, Jhuang coauthors HMDB with Hildegard Kuehne, Estíbaliz Garrote, Tomaso Poggio and Thomas Serre. The database holds around 7,000 manually annotated clips in 51 action categories, drawn from sources ranging from digitized movies to YouTube, and the team uses it to test action recognition systems under camera motion, viewpoint changes, varying video quality and occlusion.
2013

J-HMDB at Max Planck

As a member of the Perceiving Systems department at the Max Planck Institute for Intelligent Systems, Jhuang is first author of "Towards Understanding Action Recognition" at ICCV 2013, with Juergen Gall, Silvia Zuffi, Cordelia Schmid and Michael J. Black. The paper annotates human joints for the HMDB dataset, creating J-HMDB, and finds that high-level pose features greatly outperform low- and mid-level features.
2022

The Everingham Prize

At ECCV 2022, Jhuang shares the Everingham Prize with the rest of the HMDB51 dataset team, Hilde Kuehne, Estibaliz Garrote, Tomaso A. Poggio and Thomas Serre, and with the UCF101 dataset team of Khurram Soomro, Amir Roshan Zamir and Mubarak Shah.