Omnidata
收藏资源简介:
Omnidata数据集是由瑞士联邦理工学院和加州大学伯克利分校合作创建的,旨在从真实世界的3D扫描中生成多任务中层视觉数据集。该数据集包含约1450万张图像,涵盖多种视觉任务,如深度估计、表面法线估计和语义分割等。创建过程涉及使用Blender等工具对3D扫描数据进行参数化采样和渲染。Omnidata数据集的应用领域广泛,旨在解决计算机视觉中的多任务学习问题,提高模型在真实世界场景中的性能。
The Omnidata Dataset was co-created by ETH Zurich and the University of California, Berkeley, aiming to generate a multi-task mid-level vision dataset from real-world 3D scans. It contains approximately 14.5 million images covering a wide range of visual tasks such as depth estimation, surface normal estimation, semantic segmentation and more. The dataset creation process involves parametric sampling and rendering of 3D scan data using tools such as Blender. The Omnidata Dataset has broad application scenarios, and is designed to solve multi-task learning problems in computer vision and improve the performance of models in real-world scenarios.




