Pano3D

Pano3D是从球形全景图进行深度估计的新基准。其目标是以一致和全面的方式推动这项任务的进展。Pano3D 360深度估计基准提供了标准的Matterport3D训练和测试拆分，以及用于测试和训练的辅助GibsonV2分区。后者用于零拍跨数据集传输性能评估，并将其分解为3个不同的拆分，每个拆分都集中在特定的泛化轴上。

Pano3D is a novel benchmark for depth estimation from spherical panoramic images. Its goal is to advance this task in a consistent and comprehensive manner. The Pano3D 360° depth estimation benchmark provides standard training and test splits of Matterport3D, as well as auxiliary GibsonV2 partitions for both training and testing. The latter is utilized for zero-shot cross-dataset transfer performance evaluation, and it is divided into three distinct splits, each focusing on a specific generalization axis.

搜集汇总

背景与挑战

背景概述

Pano3D是一个专注于从球形全景图进行深度估计的基准数据集，包含Matterport3D和GibsonV2的标准训练与测试拆分，旨在促进360度深度估计技术的研究与发展。该数据集由希腊研究和技术中心与马德里理工大学于2021年发布。