ALTO: Alternating Latent Topologies for Implicit 3D Reconstruction

Zhen Wang; Shijie Zhou; Jeong Joon Park; Despoina Paschalidou; Suya You; Gordon Wetzstein; Leonidas J. Guibas; Achuta Kadambi

doi:10.1109/cvpr52729.2023.00033

CVPR2023顶会

ALTO: Alternating Latent Topologies for Implicit 3D Reconstruction

Zhen Wang, Shijie Zhou, Jeong Joon Park, Despoina Paschalidou, Suya You, Gordon Wetzstein, Leonidas J. Guibas, Achuta Kadambi

DOI 出版方

2023年份

12顶会引用

摘要

a) Ground Truth (b) Input (10,000 points) (c) POCO [4] (36.8 sec) (d) ALTO (10.5 sec) Figure 1. Rethinking latent topologies for fast and detailed implicit 3D reconstructions. Recent work (POCO CVPR'22 [4]) has used latent encodings for each point to preserve 3D detail. We introduce ALTO, which can alternate between latent topologies like grid latents and point latents to speed up inference and recover more detail, like the 3D reconstruction of a thin lamp-post. Scene from [52].