henry123-boy/SpaTracker

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1057 stars · 43 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

SpatialTracker is the official PyTorch implementation of a CVPR 2024 Highlight paper that tracks any 2D pixels in 3D space from RGB or RGBD video. It combines monocular depth estimation with pixel-wise 3D tracking, and a newer SpatialTrackerV2 model jointly produces depth, camera poses, and 3D tracks.

Use cases

• track pixels in 3d space from a video
• estimate 3d point trajectories from rgb video
• jointly recover depth camera pose and 3d tracks from video
• run dense point tracking on gpu for research
• lift 2d video tracking into 3d using monocular depth
• reproduce cvpr 2024 spatialtracker paper results

When to choose

• you need research-grade 3D pixel tracking from monocular video
• you want to build on or compare against state-of-the-art 3D tracking methods
• you have a CUDA GPU with ~22GB memory for dense tracking

When to avoid

• you need a production-ready, licensed product (license is non-standard)
• you lack a CUDA GPU or cannot install heavy PyTorch and depth-estimation dependencies
• you only need simple 2D point tracking without 3D lifting

Facets

library · maturity active

computer-vision machine-learning video-processing image-processing computer-vision machine-learning deep-learning python point-tracking 3d-tracking pytorch research-code cvpr-2024 depth-estimation inference video linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="henry123-boy/SpaTracker")