HaSPeR

该数据集包含了8,340张手影戏图片，分为11个类别，这些图片是从专业和业余视频中提取的。为确保数据质量，该数据集经过了人工审核，并且训练集和验证集的比例为55:45。规模上，该数据集覆盖了11个类别的8,340张图片，其任务是对手影戏进行识别。This dataset contains 8,340 hand shadow play images, categorized into 11 classes. These images are extracted from both professional and amateur videos. To ensure data quality, the dataset has undergone manual review. The split ratio of the training set to the validation set is 55:45. Overall, this dataset covers 8,340 images across 11 categories, with the core task being hand shadow play recognition.

搜集汇总

背景与挑战

背景概述

HaSPeR是一个用于手影识别任务的图像数据集，包含15种动物类别的图像，分为训练集和验证集。该数据集源自ICCV 2025 Workshop的学术论文，专门用于手影姿态识别研究，采用非商业共享许可协议。