naver/dust3r

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

7288 stars · 765 forks observed · 2026-08-28

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

DUSt3R is the official PyTorch implementation of a CVPR 2024 model that performs dense, unconstrained stereo and multi-view 3D reconstruction from arbitrary image collections without requiring camera calibration or pose priors. It regresses pointmaps with a Transformer architecture and includes a global alignment strategy, training code, and an interactive demo.

Use cases

• reconstruct a 3D model from a pair of photos
• estimate depth maps without camera calibration
• recover camera poses from unordered image collections
• run multi-view stereo reconstruction in the wild
• match pixels between images via 3D pointmaps
• train a 3D vision model on custom datasets

When to choose

• you need 3D reconstruction from images without known camera intrinsics or poses
• you want a research-grade model for depth, pose, and pointmap estimation
• you are doing robotics, SLAM, or SfM research and want a strong pretrained baseline

When to avoid

• you need a commercially licensed solution (code is CC BY-NC-SA 4.0, non-commercial only)
• you need real-time reconstruction on CPU or edge hardware
• you need a production-ready end-user application rather than a research codebase

Facets

library · maturity active

computer-vision image-processing machine-learning deep-learning gpu-computing computer-vision machine-learning artificial-intelligence robotics graphics python 3d-reconstruction pointmaps multi-view-stereo depth-estimation camera-pose-estimation transformer non-commercial-license research-code linux gpu docker

Member repositories

For agents

markdown · JSON · MCP: product_card(name="naver/dust3r")