大型低光模拟数据集
收藏资源简介:
本研究构建了一个大型低光模拟数据集,用于支持低光图像增强研究。该数据集包含22,656张图像,通过精心设计的低光模拟策略合成,旨在提供比现有数据集更大和更多样化的数据。数据集中的图像涵盖了多种场景和光照条件,适用于训练基于学习的方法。通过此数据集,研究者可以训练出能够区分曝光不足区域和正常曝光区域的模型,以及区分噪声和真实纹理的模型,从而有效提升低光图像的亮度和去噪效果。此外,数据集还提供了噪声分布图和曝光图,可作为监督信息,进一步提高训练模型的性能。
This study constructs a large-scale low-light simulation dataset to support research on low-light image enhancement. This dataset contains 22,656 images synthesized via a carefully designed low-light simulation strategy, aiming to provide larger and more diverse data than existing datasets. The images in this dataset cover various scenarios and lighting conditions, making them suitable for training learning-based methods. With this dataset, researchers can train models capable of distinguishing under-exposed regions from normally exposed regions, as well as differentiating noise from real textures, thereby effectively enhancing the brightness adjustment and denoising performance of low-light images. Additionally, the dataset also provides noise distribution maps and exposure maps, which can serve as supervision information to further improve the performance of the trained models.




