Full, unabridged title (which adds something important!):

"SAMURAI: Adapting Segment Anything Model for Zero-Shot Visual Tracking with Motion-Aware Memory"

It's the memory part that I find so impressive in the demo videos!