google-deepmind/tapnet

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

2217 stars · 212 forks observed · 2026-10-15

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

Google DeepMind's official repository for Tracking Any Point (TAP), containing the TAP-Vid and TAPVid-3D benchmarks, the TAPIR and TAPNext point-tracking models, and the RoboTAP robotics extension. It provides models and datasets for tracking arbitrary query points through video sequences.

Use cases

• track any point through a video
• evaluate point tracking models on TAP-Vid benchmark
• run TAPIR or TAPNext for video point tracking
• use point tracks for robot manipulation imitation learning
• benchmark 3D point tracking on real-world videos
• train point tracking models with bootstrapped video data

When to choose

• you need state-of-the-art point tracking models like TAPIR or TAPNext
• you want to benchmark point tracking against TAP-Vid or TAPVid-3D
• you're building robotics manipulation systems that rely on point tracks
• you need pretrained checkpoints for video motion analysis

When to avoid

• you need general object detection or segmentation rather than point tracking
• you want a production-ready application rather than research code
• you can't run GPU-accelerated deep learning models
• you need real-time tracking on edge devices without significant optimization

Facets

library · maturity active

computer-vision machine-learning deep-learning benchmarking computer-vision deep-learning robotics machine-learning python cross-platform point-tracking tapir tapnext video-understanding jax benchmark robotics trajectory-estimation gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="google-deepmind/tapnet")