A post introduces an object-focused approach to representing motion in video
The announcement calls it “localized context-aware motion representations,” contrasting it with current video representations that it says entangle motion across a scene.
TLDR
A post introduces “localized context-aware motion representations” for understanding and generating motion at the level of individual objects. It argues that current video representations entangle motion globally—a poor fit for real-world scenes where many objects move at once.
Combined views
5.9K
1 Source, first seen 18d ago
likes