vye16/shape-of-motion

Health v2 · maintenance only

Adoption not part of the score

1302 stars · 97 forks observed · 2026-08-28

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

Shape of Motion is a Python research codebase for 4D reconstruction of dynamic scenes from a single monocular video, based on the ICCV 2025 paper of the same name. It fits Gaussian splatting representations (including 2D Gaussian splatting) to video using depth maps, object masks, camera estimates, and 2D tracks.

Use cases

• reconstruct a 4D dynamic scene from a single video
• train a gaussian splatting model on the iPhone or Nvidia datasets
• fit a dynamic scene model to a custom video with estimated cameras and depths
• evaluate novel view synthesis on monocular video sequences
• use 2D gaussian splatting for better scene geometry

When to choose

• you need 4D/dynamic scene reconstruction from monocular video
• you want a research baseline for gaussian splatting on dynamic scenes
• you have GPU resources and can run preprocessing pipelines for depth, masks, and tracks

When to avoid

• you need a production-ready or real-time reconstruction system
• you lack a GPU or cannot run heavy preprocessing dependencies
• you only need static (non-dynamic) scene reconstruction, where simpler splatting tools suffice

Facets

library · maturity active

computer-vision image-processing machine-learning graphics simulation computer-vision graphics machine-learning python 4d-reconstruction gaussian-splatting monocular-video scene-reconstruction depth-estimation research-code research linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="vye16/shape-of-motion")