AGORA 姿态估计数据集

AGORA 全称 Avatars in Geography Optimized for Regression Analysis，是一个合成人类数据集，具有高真实度和非常准确的 ground truth 。此数据集包括约 14k 张训练图像和 3k 张测试图像，使用基于图像的 lighting 或渲染 3D 环境，使得每张图像中包括 5-15 人。 AGORA 数据集总共包括 173k individual person crop，提供 SPL/SMPL-X 参数及图像中每个 subject 的分割 mask 。AGORA, whose full name is Avatars in Geography Optimized for Regression Analysis, is a synthetic human dataset featuring high realism and highly accurate ground truth. This dataset includes approximately 14k training images and 3k test images. It utilizes image-based lighting or rendered 3D environments, with each image containing 5 to 15 human subjects. The AGORA dataset contains a total of 173k individual person crops, and provides SPL/SMPL-X parameters as well as segmentation masks for each subject in the images.

搜集汇总

背景与挑战

背景概述

AGORA是一个高真实度的合成人类数据集，主要用于姿态估计研究，包含约14k训练图像和3k测试图像，每张图像包含5-15人，总计173k个人物裁剪，提供SPL/SMPL-X参数和分割mask。