3D Gaussian Splatting Delivers Real-Time 1080p Scene Rendering With 2.7x Faster Training and VR Support

3D Gaussian Splatting revolutionizes real-time scene rendering, achieving stunning 1080p visuals at 30fps while delivering 2.7x faster training speeds and cutting-edge VR support through OpenXR integration.

• Researchers introduce 3D Gaussian Splatting, a novel real-time radiance field rendering method that achieves state-of-the-art visual quality at 1080p resolution with at least 30 fps by representing scenes with 3D Gaussians instead of costly neural networks.
• The open-source implementation, hosted on GitHub with over 23,600 stars, includes a PyTorch-based optimizer, real-time and network viewers built on the SIBR framework, and tools for converting custom image datasets into trainable formats using COLMAP.
• Recent updates add training speed acceleration delivering up to 2.7x faster training, along with new features including depth regularization, exposure compensation, anti-aliasing via EWA filtering, and OpenXR support for VR viewing.