SiMHand
收藏资源简介:
SiMHand数据集是由东京大学研究团队构建的一个大规模手部图像数据集,包含超过200万张从人类中心视频(如Ego4D和100DOH)中提取的手部图像。该数据集通过采用现有的手部检测器和2D手部姿态估计器,专注于挖掘具有相似手部姿态的非同款样本,以用于3D手部姿态估计的预训练。SiMHand数据集的特点是规模庞大,远远超过之前相关工作的数据集规模,能够为3D手部姿态估计任务提供丰富的训练样本。
The SiMHand dataset is a large-scale hand image dataset developed by a research team at The University of Tokyo. It contains over 2 million hand images extracted from egocentric videos such as Ego4D and 100DOH. Leveraging existing hand detectors and 2D hand pose estimators, this dataset focuses on mining non-identical samples with similar hand poses, aiming to support pre-training for 3D hand pose estimation tasks. A key characteristic of SiMHand is its exceptional scale, which far exceeds that of datasets used in previous related works, thus providing rich training samples for 3D hand pose estimation tasks.

- 1SiMHand: Mining Similar Hands for Large-Scale 3D Hand Pose Pre-training东京大学 · 2025年



