Night-haze: hazy images dataset with localized light sources for benchmarking of dehazing methods

Here we proposed two datasets for benchmarking single-image dehazing (haze removal) methods in both day and night conditions: "Night-haze" and "Night-haze-ext". "Night-haze" features: - real hazy, haze-free images and its corresponding depth maps, taken indoors; - 2 scenes – one with simpler objects and the other with more complex objects and with the presences of localized light sources; - 4 haze density levels – from absent to heavy haze; - 4 lighting levels – from sufficient to weak lighting (which are try to simulate day and nighttime lighting); - 32 images and depth maps in total. "Night-haze-ext" has similar features, but: - increased depth of scenes; - increased the number of haze density levels to 8; - thermal images in addition to regular (visible) images and depth maps; - 64 images and depth maps, 63 thermal images in total. Folders description: - night-haze: - jpg: visible images, collected using Canon 2000d in .jpg format; - raw: visible images, collected using Canon 2000d in .cr2 format; - kinnect: files, collected using Microsoft Kinnect v2; - color: visible images in .npy* format; - depth: depth map in .npy* format; - realsense: files, collected using Intel RealSence d435i; - color: visible images in .npy* format; - depth: depth map in .npy* format; - night-haze-ext: similar to night-haze, but: - kinnect: files in .png format in "color" and "depth" folders; - realsense: files in .png format in "color" and "depth" folders; - infrared: files, collected using Flir C2; - color: visible images in .png format; - spectrum: infrared images in .png format; * - numpy’ arrays format (https://numpy.org/doc/stable/reference/generated/numpy.load.html) Details about our motivation, descriptions of datasets collection processes as well as some experimental results can be found in the articles: - Night-haze: "Filin, A., Kopylov, A., Seredin, O., & Gracheva, I. (2022, July). Hazy images dataset with localized light sources for experimental evaluation of dehazing methods. In The 6th International Workshop on Deep Learning in Computational Physics (p. 19)." (https://pos.sissa.it/429/019/pdf) - Night-haze-ext: article in progress. Please, use night-haze.bib and night-haze-ext.bib (see Files section) for the corresponding references.