Abstract: This paper describes georeference contrastive learning of visual representation (GeoCLR) for efficient training of deep-learning convolutional neural networks (CNNs). The method leverages ...