facebookresearch/co-tracker

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

5080 stars · 385 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

CoTracker is a transformer-based model from Meta AI and Oxford VGG that jointly tracks any point (pixel) across a video, handling occlusions and long-term tracking. It provides pretrained checkpoints, training code, a pseudo-labeling pipeline, and demo notebooks (Colab, Hugging Face Space).

Use cases

• track any pixel through a video
• track a quasi-dense grid of points jointly across frames
• track points through occlusions and out-of-view motion
• run online (causal) point tracking on video streams
• train or fine-tune a point tracking model with pseudo-labelled real videos
• use point tracking for downstream 3D reconstruction or motion analysis

When to choose

• you need state-of-the-art point tracking in videos with occlusion handling
• you want to jointly track many points rather than independently
• you need pretrained checkpoints plus training code for research
• you want a fast transformer-based tracker suitable for online use

When to avoid

• you only need simple frame-to-frame optical flow rather than long-term point tracks
• you need a production-ready application with a UI rather than a research model
• you lack GPU resources for inference or training
• you need a permissively licensed model for commercial use without checking the custom license

Facets

library · maturity active

computer-vision machine-learning deep-learning video-processing computer-vision machine-learning artificial-intelligence python cross-platform point-tracking optical-flow transformer video-understanding pytorch research-model meta-ai video gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="facebookresearch/co-tracker")