The RobotriX

The RobotriX是一个极其逼真的大型室内数据集，专注于机器人视觉问题的深度学习应用。该数据集包含超逼真的室内场景，由机器人代理探索并与物体进行视觉上逼真的交互。数据集由Unreal Engine渲染，包含38个语义类别的800万张静态图像，记录帧率超过60帧/秒，分辨率为全高清。每帧提供RGB-D和3D信息，并附带完整标注。数据集的创建过程涉及使用虚拟现实技术捕捉人类操作员的视线和动作，以控制机器人并记录场景信息。The RobotriX数据集适用于大规模数据驱动的2D和3D机器人视觉任务研究，旨在解决机器人视觉领域的多种问题，如场景理解、深度估计和物体交互等。

The RobotriX is an ultra-realistic large-scale indoor dataset tailored for deep learning applications focused on robotic vision problems. It contains ultra-realistic indoor scenes, where robot agents explore the environments and interact with objects in visually realistic manners. Rendered using Unreal Engine, the dataset includes 8 million static images spanning 38 semantic categories, with a recording frame rate exceeding 60 frames per second and full high-definition resolution. Each frame provides RGB-D and 3D information, paired with comprehensive annotations. The dataset creation process leverages virtual reality technology to capture the gaze and motion of human operators, who control the robots and record scene data. The RobotriX dataset is applicable to large-scale data-driven 2D and 3D robotic vision task research, and aims to address diverse challenges in the robotic vision field, including scene understanding, depth estimation, object interaction, and more.