Annotated Flickr dataset for identification of professional photographers

The dataset consists of two collections, namely, users' features and photos' features. Specifically, we have collected 2,647,928 photos uploaded by 27,516 unique users to Flickr, with an average of 96 photos per user and 100 photos for more than 90% of users. Apart from the data directly collected from Flickr, the dataset has been enriched with new features that allow the analysis performed in the associated research article. One of the most valuable features of this data collection is that each photo has three Image Quality Assessment scores representing aesthetic and technical aspects. For this, we used Convolutional Neural Networks trained with human-labeled data. Furthermore, we added labels to indicate whether the user is a professional photographer, so the data are specially prepared for supervised training.

该数据集由两个集合构成，即用户特征和照片特征。具体而言，我们收集了2,647,928张由27,516位独立用户上传至Flickr的照片，平均每位用户上传96张照片，超过90%的用户上传了100张以上。除了直接从Flickr收集的数据外，数据集还通过新增特征得到了丰富，这些特征使得关联研究论文中的分析成为可能。该数据收集中最具价值的特点之一是，每张照片均拥有三个代表美学和技术方面的图像质量评估分数。为此，我们采用了基于人类标注数据的卷积神经网络进行训练。此外，我们还添加了标签以指示用户是否为专业摄影师，从而使得数据特别适用于监督训练。