This Fixed 3D Gaussian Splatting's Biggest Problems!

I sat down with Sanja Fidler, head of NVIDIA’s AI research lab in Toronto, to dive into two groundbreaking advancements: 3D Gaussian Ray Tracing and 3D Gaussian Unscented Transforms. 🚀 These innovations let you: Use distorted lenses (like fisheye and rolling shutter) for accurate 3D reconstructions Add shadows, reflections, and photorealistic lighting into your Gaussian scenes Run reconstructions with consumer RTX GPUs (or even Jetson devices!) PLUS! We talked about her latest research into diffusion based models in 3D Gaussian Splatting via Difix3D+. Whether you’re a 3D artist, AI researcher, or builder of spatial computing tools—this is the next frontier in photoreal 3D generation. 🔗 Links & Resources 📄 3D Gaussian Ray Tracing Project: https://gaussiantracer.github.io/ 📄 3D Gaussian UnscentedTransforms Project: https://research.nvidia.com/labs/toronto-ai/3DGUT/ 📄 Difix3D+ Project: https://research.nvidia.com/labs/toronto-ai/difix3d/ 🔓 3DGRUT Code (Open Source): https://github.com/nv-tlabs/3dgrut 👩‍🔬 Follow Sonja’s Lab 's Work: https://research.nvidia.com/labs/toronto-ai/ 👍 Like this interview? ✅ Subscribe for more 3D Gaussian Splatting content and deep dives into the future of 3D, AI, and spatial computing. #GaussianSplatting #3DReconstruction #NVIDIA #RayTracing #DiffusionModels #ComputerVision #GTC2025 #CosmosAI #Omniverse