Presentation
Gradient-Weighted Feature Back-Projection: A Fast Alternative to Feature Distillation in 3D Gaussian Splatting
DescriptionWe propose a training-free method for feature field rendering in 3D Gaussian Splatting, enabling fast and scalable embedding of high-dimensional features into 3D scenes. Unlike training-based feature distillation methods, which are computationally expensive and often yield feature embeddings that poorly reflect the rendered semantics, our approach back-projects 2D features onto pre-trained 3D Gaussians using influence weights derived from the rendering equation. This projection produces a queryable 3D feature field, validated on tasks including 2D and 3D segmentation, affordance transfer, and identity encoding, spanning queries using language, pixel, and synthetic embeddings. These capabilities, in turn, enable downstream applications in augmented and virtual reality, interactive scene editing, and robotics. Across different tasks, our method achieves performance comparable to or better than training-based approaches, while significantly reducing computational cost. The project page is at https://jojijoseph.github.io/3dgs-backprojection
Event Type
Technical Papers
TimeThursday, 18 December 202510:40am - 10:50am HKT
LocationMeeting Room S426+S427, Level 4