Sand-dust Image Reconstruction Benchmark (SIRB)
收藏资源简介:
Sand-dust Image Reconstruction Benchmark (SIRB) 是由河北工业大学创建的大型沙尘图像数据集,包含16000个合成沙尘图像和230个真实世界沙尘图像,覆盖多种沙尘暴退化场景。数据集通过模拟沙尘图像分布创建,用于训练卷积神经网络(CNNs)和评估沙尘图像重建算法的性能。SIRB采用全参考和无参考策略进行算法性能的定性和定量评估,为当前沙尘图像增强方法的优缺点提供了全面的见解。此外,通过在SIRB上训练的经典图像变换CNN模型作为基线,与现有沙尘去除方法进行比较,为未来基于数据驱动的沙尘去除算法研究提供了建设性的启示。
Sand-dust Image Reconstruction Benchmark (SIRB) is a large-scale sand-dust image dataset developed by Hebei University of Technology. It contains 16,000 synthetic sand-dust images and 230 real-world sand-dust images, covering diverse sandstorm degradation scenarios. The dataset is constructed by simulating the distribution of sand-dust images, and is designed for training convolutional neural networks (CNNs) and evaluating the performance of sand-dust image reconstruction algorithms. SIRB adopts both full-reference and no-reference strategies to conduct qualitative and quantitative assessments of algorithm performance, providing comprehensive insights into the strengths and weaknesses of current sand-dust image enhancement methods. Furthermore, taking the classic image transformation CNN model trained on SIRB as a baseline and comparing it with existing sand-dust removal methods offers constructive implications for future data-driven research on sand-dust removal algorithms.




