Core Concepts
CoLo-CAM verbessert die Objektlokalisierung in Videos durch Co-Lokalisierung und Farbähnlichkeiten.
Stats
"Extensive experiments on two challenging YouTube-Objects datasets of unconstrained videos show the merits of our CoLo-CAM method."
"Our method yields a single model with parameters {θ′, θ} performing both tasks, classification and localization, while handling multiple classes at once."