Nerfstudio

Nerfstudio is a 3D reconstruction framework for researchers and developers turning photos or phone video into scenes they can view from new angles. It supports model training, testing and inspection within one Python framework.

• Scene reconstruction: Train neural radiance fields (NeRFs) with Nerfacto, Instant-NGP, Mip-NeRF or TensoRF, or use Splatfacto for Gaussian splatting.
• Capture workflows: Process your own captures with COLMAP, Polycam or Record3D to build a 3D representation from images or phone video.
• Interactive viewing and exports: Inspect scenes as training progresses, revisit trained models, create camera paths for rendered videos and export point clouds.
• Research extensions: Replace model components or add methods, with profiling, benchmarking and TensorBoard or Wandb logging for comparing experiments.

Extensions connect scene workflows to Blender, Autodesk Maya and Unreal Engine. Third-party methods include Instruct-NeRF2NeRF for instruction-based scene editing and LERF for language-based scene representations.

Nerfstudio runs on your own machine or a remote server and requires an NVIDIA GPU with CUDA. Docker deployment is available. The code is open source under Apache 2.0.