Hazy Video and Image Dataset (HazyVid): A New Resource for Real World Video Dehazing Research

Hazy conditions pose significant challenges to computer vision tasks, especially in real-world video applications. While video dehazing algorithms have gained increasing attention, their development and evaluation remain hindered by the lack of large-scale, real-world hazy video datasets. To address this gap, the HazyVid is introduced, a dataset created from real-world footage captured under diverse environmental conditions. Unlike existing datasets, HazyVid does not rely on ground-truth clear videos, making it more representative of real-world haze scenarios. The dataset acquisition process is detailed, its key characteristics are analyzed, and its utility is demonstrated through benchmark experiments on video dehazing algorithms. The HazyVid dataset consists of 153 videos ranging from 1 to 15 seconds and 65,078 images, categorized into five scene types: Indoor, Mountains, Night, Road, and Rural Areas. This dataset serves as a valuable benchmark for advancing dehazing research and improving the robustness of real-world computer vision applications.