Gaussian Splat Relighting enters beta for embodied AI training
Niantic Spatial says the tool generates lighting, weather and time-of-day variants while preserving the underlying geometry, navigation and semantic data.
TLDR
Niantic Spatial introduced Gaussian Splat Relighting in beta to help train embodied AI for changing real-world conditions. The company says it uses aligned meshes to simulate new illumination, then transfers it onto Gaussian splats—creating lighting, weather and time-of-day variants while preserving the underlying geometry, navigation and semantic data.
Combined views
496
1 Source, first seen 17h ago
Gaussian Splat Relighting enters beta for embodied AI training
Niantic Spatial says the tool generates lighting, weather and time-of-day variants while preserving the underlying geometry, navigation and semantic data.
TLDR
Niantic Spatial introduced Gaussian Splat Relighting in beta to help train embodied AI for changing real-world conditions. The company says it uses aligned meshes to simulate new illumination, then transfers it onto Gaussian splats—creating lighting, weather and time-of-day variants while preserving the underlying geometry, navigation and semantic data.