UnrealEgo

我们提出了UnrealEgo，即一个新的大规模自然主义数据集，用于以自我为中心的3D人类姿势估计。UnrealEgo基于配备两个鱼眼摄像头的高级眼镜概念，可在不受限制的环境中使用。我们设计了它们的虚拟原型，并将它们附加到3D人体模型上以进行立体视图捕获。接下来，我们将生成大量的人类运动。因此，UnrealEgo是第一个提供现有以自我为中心的数据集中的运动种类最多的野外立体图像的数据集。此外，我们提出了一种新的基准方法，该方法具有简单而有效的思想，即为立体声输入设计2D关键点估计模块以改善3D人体姿势估计。广泛的实验表明，我们的方法在定性和定量上都优于以前的最新方法。We introduce UnrealEgo, a novel large-scale naturalistic dataset for egocentric 3D human pose estimation. UnrealEgo is built upon the concept of advanced glasses equipped with two fisheye cameras, enabling deployment in unconstrained environments. We designed virtual prototypes of these glasses and attached them to 3D human body models for stereoscopic view capture. Subsequently, we generated a large corpus of diverse human motion sequences. As a result, UnrealEgo is the first stereoscopic image dataset that offers the most abundant variety of motions among existing egocentric datasets captured in the wild. Furthermore, we propose a novel baseline method with a simple yet effective insight: designing a 2D keypoint estimation module for stereo inputs to enhance 3D human pose estimation. Extensive experiments demonstrate that our method outperforms previous state-of-the-art approaches in both qualitative and quantitative aspects.

搜集汇总

背景与挑战

背景概述

UnrealEgo是一个大规模自然主义数据集，专为以自我为中心的3D人体姿势估计设计，基于双鱼眼摄像头在无约束环境中捕获立体图像，提供了最丰富的运动种类。该数据集还提出了一种新的基准方法，通过优化2D关键点估计来提升3D姿势估计性能，实验证明其优于现有方法。