Voxel51/hand-keypoints

这是一个包含846个样本的FiftyOne数据集，其中样本是从MPII人体姿态数据集和新西兰手语练习数据集中提取的手部RGB图像，并进行了手动标注。数据集包含2800只手上的21个关键点的2D位置信息，分为训练集2000只手和测试集800只手。该数据集被用于训练和评估单张图像中手部关键点检测方法，并作为评估其单视图2D手部关键点检测器准确性的基准。数据集来源于YouTube视频和展示新西兰手语手势的图像，用于评估检测器的泛化能力。

This is a FiftyOne dataset with 846 samples, consisting of manually annotated RGB images of hands sourced from the MPII Human Pose dataset and the New Zealand Sign Language (NZSL) Exercises. The dataset includes 2D locations for 21 keypoints on 2800 hands, split into a training set of 2000 hands and a testing set of 800 hands. It was used for training and evaluating hand keypoint detection methods for single images, serving as a benchmark to assess the accuracy of their single image 2D hand keypoint detector. The images are from YouTube videos depicting everyday human activities (MPII) and images showing a variety of hand poses from people using NZSL, which were used to evaluate the generalization capabilities of the detector.