UASOL (A large-scale high-resolution outdoor stereo dataset)

UASOLis一种新的数据集，用于从单个和立体声RGB图像中估计室外深度。数据集是从行人的角度获取的。当前，最新颖的方法利用了基于深度学习的技术，这些技术已被证明优于传统的最先进的计算机视觉方法。尽管如此，这些方法仍需要大量可靠的地面真实数据。尽管已经存在几个可用于深度估计的数据集，但从自我中心的角度来看，几乎没有一个数据集是面向室外的。我们的数据集从人的角度引入了大量高清对的彩色帧和相应的深度图。此外，所提出的数据集还具有人类互动和数据的巨大可变性。

UASOL is a novel dataset for outdoor depth estimation from monocular and stereo RGB images. It is captured from a pedestrian's first-person perspective. Currently, state-of-the-art depth estimation methods rely on deep learning-based techniques, which have been demonstrated to outperform traditional state-of-the-art computer vision approaches. Nevertheless, these methods require large quantities of reliable ground-truth data. Although several depth estimation datasets have already been developed, few are tailored for outdoor environments from an egocentric viewpoint. Our dataset offers a large corpus of high-definition color frames paired with corresponding depth maps, all collected from a human's perspective. Furthermore, the proposed dataset features significant variability in human interactions and data samples.

搜集汇总

背景与挑战

背景概述

UASOL是一个大规模高分辨率室外立体数据集，专为从单目和立体RGB图像估计深度而设计。该数据集从行人视角采集，提供了大量高清彩色帧及其对应的深度图，并包含人类互动和高度可变的数据。