GSN: Generalisable Segmentation in Neural Radiance Field
This work addresses the need for efficient and generalizable scene understanding in computer vision, though it is incremental as it builds on existing generalizable RF methods.
The authors tackled the problem of enabling multi-view semantic segmentation in arbitrary new scenes using generalizable neural radiance fields, achieving results on par with traditional scene-specific methods and significantly closing the performance gap.
Traditional Radiance Field (RF) representations capture details of a specific scene and must be trained afresh on each scene. Semantic feature fields have been added to RFs to facilitate several segmentation tasks. Generalised RF representations learn the principles of view interpolation. A generalised RF can render new views of an unknown and untrained scene, given a few views. We present a way to distil feature fields into the generalised GNT representation. Our GSN representation generates new views of unseen scenes on the fly along with consistent, per-pixel semantic features. This enables multi-view segmentation of arbitrary new scenes. We show different semantic features being distilled into generalised RFs. Our multi-view segmentation results are on par with methods that use traditional RFs. GSN closes the gap between standard and generalisable RF methods significantly. Project Page: https://vinayak-vg.github.io/GSN/