DAWN: Dual Augmented Memory Network for Unsupervised Video Object Tracking

Project Page

Overview

This webpage contains visual demostration for paper DAWN: Dual Augmented Memory Network for Unsupervised Video Object Tracking. There will be a comparison between DAWN and other state-of-art trackers, especially, C-COT the most precise tracker of VOT 2016 and 2017, and DaSiam the champion of VOT 2018 real-time competition.

overview

Comparison with other trackers

Compared with other state-of-art trackers, DAWN is robust against partial and total occlusion (OC), severe motion blur (MB), abrupt changes in target appearance (AC), multiple object instances, and similar foreground and background features (SB).


basketball: OC, SB

butterfly: AC, SB

girl: OC

bolt1: SB

ball2: MB, SB

fish2: AC, SB

frisbee: OC, AC

zebrafish1: AC, SB

godfather: OC, SB

gymnastics1: MB, SB

helicopter: AC

motocross2: MB, AC

soccer2: OC, MB

wiper: OC, MB

ants1: SB