DarkVision
收藏资源简介:
DarkVision是由清华大学创建的多光照、多相机、低光数据集,旨在服务于图像增强和目标检测任务。该数据集包含900个静态场景和32个动态场景,捕捉了15个类别的物体,每个场景使用三种不同级别的相机在5种光照条件下拍摄。静态场景图像和动态视频分别包含约7,344和320,667个实例。DarkVision通过提供像素级对齐的亮暗图像对,为图像/视频增强和目标检测建立了基准。数据集的应用领域包括夜间监控、摄影、高速摄影和自动驾驶等,旨在解决低光环境下的视觉感知问题。
DarkVision is a low-light dataset with multi-illumination and multi-camera setups developed by Tsinghua University, targeting image enhancement and object detection tasks. This dataset includes 900 static scenes and 32 dynamic scenes, covering objects from 15 categories. Each scene is captured under 5 illumination conditions using three distinct camera setups. The static scene images and dynamic videos contain approximately 7,344 and 320,667 instances respectively. DarkVision establishes benchmarks for image/video enhancement and object detection by providing pixel-aligned bright-dark image pairs. Its application fields cover night surveillance, photography, high-speed photography, autonomous driving and other scenarios, aiming to address visual perception challenges in low-light environments.

- 1DarkVision: A Benchmark for Low-light Image/Video Perception清华大学 · 2023年



