VIsual PERception benchmark (VIPER)

VIPER数据集是由达姆施塔特工业大学、澳大利亚国立大学和英特尔实验室联合创建的视觉感知基准。该数据集包含254,064个高分辨率视频帧，涵盖多种环境条件下的视觉任务，如光流、语义实例分割、物体检测和跟踪等。数据集通过在虚拟世界中驾驶、骑行和步行共184公里收集，采用新颖的方法从模拟世界中收集地面实况数据。VIPER数据集旨在支持开发能够在复杂环境中构建和维护全面环境模型的广泛能力视觉感知系统，解决现实世界中大规模数据收集的技术难题。

The VIPER dataset is a visual perception benchmark jointly created by Technische Universität Darmstadt, Australian National University, and Intel Labs. This dataset contains 254,064 high-resolution video frames, covering visual tasks under various environmental conditions such as optical flow, semantic instance segmentation, object detection, and tracking. It was collected through a total of 184 kilometers of traversal via driving, cycling, and walking within virtual worlds, adopting a novel method to gather ground-truth data from simulated environments. The VIPER dataset aims to support the development of visual perception systems with broad capabilities to construct and maintain comprehensive environmental models in complex environments, addressing the technical challenges of large-scale data collection in the real world.