NToP
收藏资源简介:
NToP(NeRF驱动的顶视图人体姿态数据集)是由开姆尼茨工业大学创建的大规模数据集,专为鱼眼相机顶视图的人体姿态估计设计。该数据集包含超过57万张高质量半合成人体图像,并附有2D和3D关键点标注。数据集的创建过程利用了NeRF技术,从现有的2D和3D数据集中生成,特别针对顶视图鱼眼视角进行了优化。NToP数据集主要应用于顶视图人体姿态估计,旨在解决现有数据集在顶视图视角下的局限性,提高神经网络在2D和3D顶视图人体姿态估计任务中的性能。
NToP (NeRF-powered top-view human pose dataset) is a large-scale dataset created by Chemnitz University of Technology, purpose-built for human pose estimation with fisheye top-view cameras. This dataset contains over 570,000 high-quality semi-synthetic human images paired with 2D and 3D keypoint annotations. The dataset was generated from existing 2D and 3D datasets leveraging NeRF technology, and specially optimized for the fisheye top-view perspective. The NToP dataset is primarily applied to top-view human pose estimation, aiming to address the limitations of existing datasets under top-view perspectives and improve the performance of neural networks in 2D and 3D top-view human pose estimation tasks.

- 1NToP: NeRF-Powered Large-scale Dataset Generation for 2D and 3D Human Pose Estimation in Top-View Fisheye Images开姆尼茨工业大学 · 2024年



