CS Team Wins at the Epic Kitchen Action Anticipation Challenge

A team of PhD students advised by Professor Yiannis Aloimonos and Professor Cornelia Fermuller achieved the first and second place over the two portions of the 2020 Epic Action Anticipation Challenge. 

Anticipating future human activity from video is an area of great importance in computer vision with applications in human-robot interaction, surveillance, navigation, etc. To motivate research towards solving this important problem, the EPIC Kitchens Action Anticipation Challenge was launched, where 20 international teams from leading companies and universities competed. 

EPIC-Kitchens is the largest dataset in first-person (egocentric) vision with multi-faceted non-scripted recordings in native environments. For example, in the wearer's homes, it captures all daily activities in the kitchen over multiple days. Annotations are collected using a novel live audio commentary approach.

The UMD team led by Eadom Dessalene including team members Michael Maynord and Chinmaya Devaraj secured first place (Unseen Kitchens) and second place (Seen Kitchen) under the Action Anticipation track . The team recently presented their work at the EPIC@CVPR2020 workshop. 

"We are very excited with these results,” said Professor Aloimonos, “because they constitute the basis for a fundamental problem, namely the problem of predicting the next action that someone (a human)  is about to do, given a video of the activity. So far, the community has been studying the problem of action recognition, i.e. recognizing an action that the system has seen before. This work goes a step further and predicts what will happen next, much like a human can do by watching someone else's activity. This is a modern problem of very high interest and this prize is a great calibration parameter for our Lab."

A preprint of the work can be found here.