News

Cybercore R&D team won 1st place in YouTube-VOS 2021 Challenge. The team will make 2 presentations during CVPR 2021, a prestigious Image AI conference.

2021.06.16

Researchers from Cybercore Co., Ltd. (Morioka, Iwate) have won the 1st place among 331 entrants and 39 finalists in the international computer vision competition YouTube-VOS 2021 Challenge. The team will have a presentation at one of the most prestigious computer vision conferences CVPR 2021 on June 20th.

Besides the achievement in the YouTube-VOS 2021 Challenge, as previously announced in May, Cybercore team has also won the 1strank in Vehicle Re-Identification in Papers with Code. This algorithm was developed during the AI City Challenge competition The team will present their solution at the CVPR 2021’s AI City Challenge section in June as well.

 

Overview of YouTube-VOS 2021 Challenge

YouTube-VOS 2021 Challenge has 3 different competitions, and we joined the Video Instance Segmentation one. Our team participated the challenge with 4 researchers from Cybercore’s Ho Chi Minh R&D center.

Video Instance Segmentation is a multi-task problem, which performs three tasks simultaneously, including detection, segmentation, and tracking. Our approach takes the advantage of data efficiency analysis, multi-task learning, and deep-supervision. This not only improves the model’s speed and memory, while yields higher accuracy and reduces training time thanks to the joined optimization framework.

Our solution has a wide range of applications, such as video understanding, object detection and tracking for autonomous driving, human behavior, or motion analysis. Taking the video understanding problem as an example, our solution can segment any objects (person, vehicles, animals) in the videos and track them for specific purposes. Segmented results can serve as another source for action analysis, such as in sport, e.g., supporting coaches with more information to train their teams. In autonomous driving applications, our solution can segment moving objects on the streets, such as pedestrians, riders, other vehicles, or traffic signs. This helps autonomous vehicles be able to take appropriate action on the ride.

 

CVPR 2021 workshop
http://cvpr2021.thecvf.com/workshops-schedule

YouTube-VOS Challenge Outline and Results
https://competitions.codalab.org/competitions/28988#results

 

Malios Bldg.10F, 2-9-1,
Moriokaekinishidori, Morioka-shi, Iwate
TEL : 019-681-8776

Copyright(c) Cybercore Co.,LTd. All Rights Reserved.