Skip to content

CoTracker

Track dense points jointly across video frames

Visit official site

What CoTracker is

CoTracker is described as a Transformer model that tracks dense points jointly in a video sequence. The material contrasts it with methods that track points independently and ignore their correlation.

What you can do with it

  • Track dense points jointly in a video sequence
  • Use virtual trajectories to follow up to 70,000 points
  • Operate causally on short windows for online tasks
  • Keep tracking points under occlusion or out of view

Application scenarios

  • Track visual objects in video
  • Handle multiple targets at once
  • Follow objects over long video sequences

Pros & cons

✓ What we like

  • Joint tracking accounts for correlations between points
  • Causal short windows suit online tasks

! What to watch out for

  • Tracking is described for video sequences
  • Virtual trajectories underpin the large point count

FAQ

How many points can it track?

The material states virtual trajectories allow joint tracking of up to 70,000 points.

What happens under occlusion?

The feature list says points can be tracked for a long time even when occluded or out of view.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review