zju3dv/street_gaussians

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1388 stars · 108 forks observed · 2026-08-28

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

Street Gaussians is a research implementation of the ECCV 2024 paper 'Modeling Dynamic Urban Scenes with Gaussian Splatting', which reconstructs dynamic street scenes from autonomous driving data using 3D Gaussian splatting. It processes Waymo Open Dataset sequences to model static backgrounds and dynamic objects as separate Gaussian representations for novel view synthesis.

Use cases

• reconstruct dynamic urban driving scenes in 3D
• render novel views of street scenes from Waymo data
• model moving vehicles and static backgrounds separately with Gaussian splatting
• research on dynamic scene reconstruction for autonomous driving
• compare against NeRF-based methods like EmerNeRF on driving benchmarks
• visualize 4D street scene reconstructions

When to choose

• you need state-of-the-art dynamic urban scene reconstruction from driving datasets
• you work with the Waymo Open Dataset and want Gaussian splatting baselines
• you are researching 3D/4D scene reconstruction or autonomous driving simulation
• you want editable, decomposed representations of dynamic objects in street scenes

When to avoid

• you need a production-ready or well-supported end-user application
• you lack a CUDA-capable GPU or cannot handle large driving datasets
• you need real-time reconstruction rather than offline training
• you want a general-purpose 3D reconstruction tool for arbitrary scenes

Facets

library · maturity active

machine-learning graphics simulation image-processing computer-vision autonomous-vehicles deep-learning graphics python gaussian-splatting 3d-reconstruction neural-rendering waymo-open-dataset urban-scenes eccv-2024 research-code linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="zju3dv/street_gaussians")