HengyiWang/spann3r

Health v2 · maintenance only

Flags: no_releases no_license

Adoption not part of the score

1141 stars · 52 forks observed · 2026-08-28

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

Spann3R is a transformer-based model for dense 3D reconstruction from ordered or unordered image collections, built on the DUSt3R paradigm. It uses an external spatial memory to regress per-image pointmaps in a global coordinate system without test-time optimization or camera pose estimation.

Use cases

• reconstruct 3d scenes from image sequences
• dense 3d reconstruction from uncalibrated photos
• incremental 3d mapping without camera pose estimation
• real-time 3d reconstruction from video frames
• reconstruct dynamic scenes from images
• export 3d reconstruction to nerfstudio

When to choose

• you need pointmap-based 3D reconstruction from images without SfM preprocessing
• you want incremental reconstruction in a global coordinate frame with a single forward pass
• you need to handle both static and dynamic scenes
• you want to feed reconstructions into Nerfstudio

When to avoid

• you need a production-ready, licensed product (license is non-standard/unclear)
• you lack a CUDA-capable GPU, since it requires PyTorch with CUDA and compiled kernels
• you need metrically accurate, survey-grade 3D reconstruction
• you need a lightweight CPU-only pipeline

Facets

library · maturity active

machine-learning deep-learning computer-vision image-processing computer-vision machine-learning deep-learning artificial-intelligence python 3d-reconstruction pointmap spatial-memory transformer dust3r nerfstudio research-code pytorch linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="HengyiWang/spann3r")