Cloud coverage in snapshot camera images

We have proposed a deep convolution neural network (CNN) approach for the accurate estimation of the cloud coverage (CC) from images captured by a consumer camera, i.e., snapshot pictures. The results have shown that the present CNN has the potential to allow consumer cameras to be used as remote weather sensors. (see the journal paper: https://doi.org/10.2151/sola.2017-043) Here we publish the dataset used for the training and evaluation of the proposed CNN. The dataset consists of camera images and their cloud coverages (integer values ranging from 0 to 10), which were labeled by human labelers. As described in the above-mentioned paper, most of the camera images were collected from the Flickr website, and some of their copyrights are not clear. Therefore, the image data themselves are not included in the dataset, and they should be downloaded from the internet. Please read the documents for how to download the image data. If you use this dataset, please read "Use Constraints" shown at "Detail info." mode.

本研究提出了一种基于深度卷积神经网络（Convolution Neural Network，CNN）的方法，可通过消费级相机拍摄的快照图像精准估算云量（Cloud Coverage，CC）。研究结果表明，本研究所提出的卷积神经网络有望使消费级相机作为远程气象传感器得以应用。（详见期刊论文：https://doi.org/10.2151/sola.2017-043）本次我们发布了用于训练及评估上述卷积神经网络的数据集。该数据集包含相机拍摄图像及其对应的云量（Cloud Coverage，CC，取值范围为0至10的整数），所有标注均由人工标注员完成。如上述论文所述，该数据集中的大部分相机图像采集自Flickr网站，且部分图像的版权归属尚不明确。因此，本数据集未包含原始图像数据，用户需自行从互联网下载。请查阅相关文档了解图像数据的下载方法。若您使用本数据集，请查阅“详细信息”模式下的“使用约束”条款。