PKU-VCL-3DV/SLAM3R

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1344 stars · 102 forks observed · 2026-08-28

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

SLAM3R is a real-time dense 3D scene reconstruction system that regresses 3D points from monocular RGB video using feed-forward neural networks, without explicitly estimating camera parameters. It is the research code release for a CVPR 2025 highlight paper, including demos, a Gradio interface, evaluation, and training pipelines.

Use cases

• reconstruct dense 3D scenes from monocular RGB video in real time
• run SLAM without estimating camera intrinsics or pose explicitly
• evaluate dense reconstruction quality on the Replica dataset
• train a feed-forward point regression model for 3D reconstruction
• visualize online 3D reconstruction from self-captured video

When to choose

• you need real-time dense 3D reconstruction from plain RGB video without depth sensors
• you want a learning-based alternative to traditional SLAM pipelines
• you are doing research on feed-forward 3D scene reconstruction

When to avoid

• you need a production SLAM system for robotics navigation with loop closure and localization guarantees
• you lack a CUDA-capable GPU
• you need LiDAR or RGB-D input support

Facets

library · maturity active

computer-vision machine-learning deep-learning simulation graphics computer-vision robotics deep-learning autonomous-vehicles python slam 3d-reconstruction monocular-video point-cloud neural-networks cvpr-2025 real-time gradio linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="PKU-VCL-3DV/SLAM3R")