Skip to content

InterTrack

Tracking human-object interaction in monocular RGB video

Visit official site

What InterTrack is

The technique is described as a tracking method for human and object interaction in monocular RGB video that stays consistent under occlusion and dynamic motion.

How it works

  • Reconstruct each frame with a single-view method
  • Predict SMPL vertices with an efficient autoencoder
  • Use temporal consistency correspondence

Model notes

  • ProciGen-Video holds 10 hours of synthetic video
  • Experiments are reported on BEHAVE and InterCap
  • Results are compared with template-based tracking

Pros & cons

✓ What we like

  • No object template is required
  • Temporal cues predict smooth object rotation under occlusion

! What to watch out for

  • Input is monocular RGB video
  • Training relies on synthetic interaction data

FAQ

Does it need an object template?

The material states the method works without any object template.

Which datasets are named?

The material names ProciGen-Video, BEHAVE and InterCap.

Last reviewed: 2026-11-15

More AI video generation tools

View all →

How we review