SHIFT, DynamicReplica, MOVi-F, PointOdyssey
收藏资源简介:
该数据集由NVIDIA Research和布朗大学的研究团队创建,主要用于单目场景流估计任务。数据集包含超过100万条样本,涵盖了室内和室外场景,数据来源包括真实场景和合成场景。数据集的创建过程通过多摄像头采集动态场景的RGB图像、深度图、光流和场景流信息,并通过伪标签生成技术补充缺失的标注信息。该数据集的应用领域包括增强现实、自动驾驶和机器人技术,旨在解决单目场景流估计中的泛化问题,提升模型在未见过的场景中的表现。
This dataset was created by a research team from NVIDIA Research and Brown University, primarily for the task of monocular scene flow estimation. It contains over one million samples covering both indoor and outdoor scenarios, with data sourced from real-world and synthetic scenes. During its creation, RGB images, depth maps, optical flow, and scene flow information of dynamic scenes were collected via multiple cameras, and missing annotation information was supplemented using pseudo-label generation techniques. The dataset can be applied in fields including augmented reality, autonomous driving and robotics, aiming to address the generalization issue in monocular scene flow estimation and improve the performance of models on unseen scenarios.




