1: \begin{abstract}

2: % \todo{(Rick) replace ``via'' with ``using'' or ``with'' in title? (Jon) agreed}

3: We present a method for reconstructing high-quality meshes of large unbounded real-world scenes suitable for photorealistic novel view synthesis. We first optimize a hybrid neural volume-surface scene representation designed to have well-behaved level sets that correspond to surfaces in the scene. We then bake this representation into a high-quality triangle mesh, which we equip with a simple and fast view-dependent appearance model based on spherical Gaussians. Finally, we optimize this baked representation to best reproduce the captured viewpoints, resulting in a model that can leverage accelerated polygon rasterization pipelines for real-time view synthesis on commodity hardware. Our approach outperforms previous scene representations for real-time rendering in terms of accuracy, speed, and power consumption, and produces high quality meshes that enable applications such as appearance editing and physical simulation.

4: % MLP-based scene representation designed to produce a well-defined output surface with a simple and fast view-dependent shading model based on spherical Gaussians. After convergence, this representation is baked out into a high resolution triangle mesh. Finally, we finetune the appearance model on top of this mesh to enable photorealistic rendering of novel views. The final result can be rendered at real-time framerates on commodity hardware.

5: \end{abstract}

6: