Kubric
收藏资源简介:
Kubric是一个由Google Research开发的开源数据集生成框架,旨在通过合成数据解决真实数据收集和标注中的难题。该框架能够生成大规模、高质量的合成数据集,支持包括3D NeRF模型到光流估计等多种视觉任务。Kubric通过与PyBullet和Blender等工具的集成,实现了对数据生成过程的精细控制,并提供了丰富的标注信息,如分割、深度、光流等。此外,Kubric还支持在数千台机器上分布式运行,生成TB级别的数据,极大地提高了数据生成的效率和规模。该数据集的应用领域广泛,包括但不限于机器学习模型的训练和评估,旨在解决数据收集和标注中的成本、隐私和法律问题。
Kubric is an open-source dataset generation framework developed by Google Research, which aims to address the challenges in real-world data collection and annotation via synthetic data. It is capable of generating large-scale, high-quality synthetic datasets that support a diverse array of visual tasks, ranging from 3D NeRF modeling to optical flow estimation. Through integration with tools including PyBullet and Blender, Kubric enables precise fine-grained control over the data generation process, while offering comprehensive annotations such as segmentation masks, depth maps, optical flow and more. Additionally, Kubric supports distributed execution across thousands of machines to produce terabyte-scale datasets, drastically enhancing both the efficiency and scale of data generation. With broad application domains covering but not limited to the training and evaluation of machine learning models, Kubric is designed to resolve the cost, privacy and legal issues associated with real-world data collection and annotation.




