데이터셋 탐색

The real captured dataset of LOL contains 500 low/normallight image pairs. Most low-light images are collected by changing exposure time and ISO, while other configurations of the cameras are fixed. We capture images from a variety of scenes, e.g., houses, campuses, clubs, streets. Since camera shaking, object movement, and lightness changing may cause misalignment between the image pairs, inspired by [41], a three-step shooting strategy is used to eliminate such misalignments between the image pairs in our dataset. For one scene, we first shoot two normal-light images $N_1$ and $N_2$. Then, we change the exposure time and ISO to capture a series of low-light images. Finally, we set the exposure time and ISO back to shoot another two normal-light images $N_3$ and $N_4$. The average of $N_i (i = 1,2,3,4)$ is treated as the ground-truth $G=\frac{1}{4}\sum^4_{i=1}N_i$. Then, we check whether there is object or camera movement. Specifically, the misalignment for these normal-light images is measured by $M=\frac{1}{4}\sum^4_{i=1}MSE(Ni, G)$. If M > 0.1, we abandon the corresponding pair. These raw images are resized to 400 × 600 and converted to Portable Network Graphics format. The dataset is publicly available.

To make synthetic images match the property of real dark photography, we analyze the illumination distribution of low-light images. We collect 270 low-light images from public MEF [42], NPE [6], LIME [8], DICM [43], VV,2 and Fusion [44] dataset, transform the imagesT into YCbCr channel and calculate the histogram of Y channel. We also collect 1000 raw images from RAISE [45] as normal-light images and calculate the histogram of Y channel in YCbCr. Raw images contain more information than the converted results. For raw images, all operations used to generate pixel values are performed in one step on the base data, making the result more accurate. 1000 raw images in RAISE [45] are used to synthesize low-light images. Interface provided by Adobe Lightroom is used and we try different kinds of parameters to make the histogram of Y channel fit the result in low-light images. Final parameter configuration can be found in the supplementary material. The illumination distribution of synthetic images matches that of low-light images. Finally, we resize these raw images to 400 × 600 and convert them to Portable Network Graphics format.