Abstract: Radiance field methods have achieved photorealistic novel view synthesis and geometry reconstruction. But they are mostly applied in per-scene optimization or small-baseline settings. While several recent works investigate feed-forward reconstruction with large baselines by utilizing transformers, they all operate with a standard global attention mechanism and hence ignore the local nature of 3D reconstruction. We propose a method that unifies local and global reasoning in transformer layers, resulting in improved quality and faster convergence.

Method: LaRa represents scenes as Gaussian Volumes and combines this with an image encoder and Group Attention Layers for efficient feed-forward reconstruction:

1. Scenes are represented as Gaussian Volume.
2. An embedding volume models 3D prior, leaned from dataset.
3. Extract per-view DINO features and lift them into 3D feature volumes.
4. Volume attention between feature volume and embedding volume, ouputs Gaussian Volume.
5. Transform Gaussian volume into coarse-to-fine 2D Gaussian primitives, using splatting for efficient rendering.