LaRa: Efficient Large-Baseline Radiance Fields
This work addresses the challenge of efficient large-baseline radiance field reconstruction for novel view synthesis and geometry, representing an incremental improvement over existing transformer-based methods by incorporating local attention.
The paper tackles the problem of feed-forward reconstruction of radiance fields from large-baseline images, which previous methods struggled with due to inefficient global attention mechanisms. The result is a method that unifies local and global reasoning, achieving high-fidelity 360-degree reconstruction with robustness in zero-shot and out-of-domain testing, trained in two days on four GPUs.
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. Our model represents scenes as Gaussian Volumes and combines this with an image encoder and Group Attention Layers for efficient feed-forward reconstruction. Experimental results demonstrate that our model, trained for two days on four GPUs, demonstrates high fidelity in reconstructing 360 deg radiance fields, and robustness to zero-shot and out-of-domain testing. Our project Page: https://apchenstu.github.io/LaRa/.