We propose a fully automatic minutiae extractor, called MinutiaeNet, based on\ndeep neural networks with compact feature representation for fast comparison of\nminutiae sets. Specifically, first a network, called CoarseNet, estimates the\nminutiae score map and minutiae orientation based on convolutional neural\nnetwork and fingerprint domain knowledge (enhanced image, orientation field,\nand segmentation map). Subsequently, another network, called FineNet, refines\nthe candidate minutiae locations based on score map. We demonstrate the\neffectiveness of using the fingerprint domain knowledge together with the deep\nnetworks. Experimental results on both latent (NIST SD27) and plain (FVC 2004)\npublic domain fingerprint datasets provide comprehensive empirical support for\nthe merits of our method. Further, our method finds minutiae sets that are\nbetter in terms of precision and recall in comparison with state-of-the-art on\nthese two datasets. Given the lack of annotated fingerprint datasets with\nminutiae ground truth, the proposed approach to robust minutiae detection will\nbe useful to train network-based fingerprint matching algorithms as well as for\nevaluating fingerprint individuality at scale. MinutiaeNet is implemented in\nTensorflow: https://github.com/luannd/MinutiaeNet\n