szymanowiczs/splatter-image

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1106 stars · 82 forks observed · 2026-08-28

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

Official PyTorch implementation of 'Splatter Image: Ultra-Fast Single-View 3D Reconstruction' (CVPR 2024), which uses an image-to-image network to predict a 3D Gaussian per pixel for monocular 3D object reconstruction at 38 FPS. Includes pretrained models for multiple datasets, a Gradio demo, and multi-GPU training support.

Use cases

• reconstruct a 3D model from a single photo
• generate 3D Gaussians from one image of an object
• render novel views of an object from a single view
• train a single-view 3D reconstruction model on Objaverse or ShapeNet
• run a fast open-category 3D reconstruction demo locally
• reconstruct video frames of an object into 3D frame-by-frame

When to choose

• you need ultra-fast (38 FPS) feed-forward single-view 3D object reconstruction
• you want pretrained models for ShapeNet, CO3D, or Objaverse categories
• you want a research baseline with SOTA results on multi-category ShapeNet
• you have an NVIDIA GPU and want Gaussian Splatting-based monocular reconstruction

When to avoid

• you need full scene reconstruction rather than single objects
• you have no CUDA-capable NVIDIA GPU
• you need watertight meshes rather than Gaussian splat representations
• you need a production-ready end-user application rather than research code

Facets

library · maturity active

machine-learning deep-learning image-processing computer-vision graphics computer-vision machine-learning graphics deep-learning python gaussian-splatting 3d-reconstruction single-view-3d cvpr-2024 research-code neural-networks 3d-gaussians gpu linux

Member repositories

For agents

markdown · JSON · MCP: product_card(name="szymanowiczs/splatter-image")