CUT3R/CUT3R

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1486 stars · 89 forks observed · 2026-08-28

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

CUT3R is the official PyTorch implementation of 'Continuous 3D Perception Model with Persistent State' (CVPR 2025 Oral), a stateful recurrent transformer that processes streams of RGB images and produces metric-scale pointmaps in a shared coordinate system. It supports online dense 3D scene reconstruction, dynamic scene reconstruction, and probing of unobserved scene regions from video or unordered photo collections.

Use cases

• reconstruct 3D scenes from a stream of RGB images
• estimate per-pixel 3D pointmaps from video frames online
• reconstruct dynamic scenes without per-video optimization
• predict depth and camera pose from image sequences
• infer unseen regions of a scene from virtual views
• run multi-view stereo reconstruction on photo collections
• evaluate 3D/4D perception models on benchmark datasets

When to choose

• you need feed-forward online 3D reconstruction from RGB-only inputs
• you want a unified model handling static and dynamic scenes
• you need metric-scale pointmaps in a common coordinate system
• you are doing research on stateful 3D perception or 4D reconstruction

When to avoid

• you need a production-ready 3D scanning pipeline with SLAM guarantees
• you lack a CUDA-capable GPU for inference
• you need a lightweight real-time mobile 3D solution
• you require permissive licensing for commercial use (license is non-standard)

Facets

library · maturity active

machine-learning deep-learning computer-vision image-processing simulation computer-vision deep-learning machine-learning artificial-intelligence python cross-platform 3d-reconstruction pointmap stateful-model transformer cvpr-2025 research-code depth-estimation pose-estimation video pytorch gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="CUT3R/CUT3R")