irelandgre.blogg.se

1, the dataset covers a broad range of domains, e.g., city scenes, family photos, nature, and historical events. In contrast to existing datasets, it contains more than one million Flickr images captured in the period from 1930 to 1999. In this paper, we introduce a novel dataset Date Estimation in the Wild and make it publicly available to support further research. For this reason, a huge dataset covering all kinds of concepts is necessary, which additionally enables the training of convolutional neural networks. Existing approaches either rely on datasets solely containing historical color images or focus on specific concepts like cities, cars, persons, or historical documents and are therefore unable to learn the temporal differences of the broad variety of motives. But date estimation is an interesting and challenging task for historians, archivists, and even for sorting (digitized) personal photo collections chronologically. However, estimating automatically the capturing time of (historical) photos has been rarely addressed yet and existing benchmark datasets do not contain enough images captured before 2000. In particular, such datasets are a prerequisite for the training of deep learning systems. In recent years, huge datasets (e.g., ImageNet , YFCC100M ) were introduced fostering research for many computer vision tasks. This process is experimental and the keywords may be updated as the learning algorithm improves. These keywords were added by machine and not by the authors. Experimental results demonstrate that these baselines are already superior to annotations of untrained humans. In addition, we propose two baseline approaches for regression and classification, respectively, relying on state-of-the-art deep convolutional neural networks. The dataset consists of more than one million images crawled from Flickr and contains a large number of different motives. In contrast to previous work, the dataset is neither restricted to color photography nor to specific visual concepts. In this paper, we introduce a novel dataset Date Estimation in the Wild for the task of predicting the acquisition year of images captured in the period from 1930 to 1999. The problem of automatically estimating the creation date of photos has been addressed rarely in the past.