A deep hashing model typically has two main learning objectives: to make the\nlearned binary hash codes discriminative and to minimize a quantization error.\nWith further constraints such as bit balance and code orthogonality, it is not\nuncommon for existing models to employ a large number (>4) of losses. This\nleads to difficulties in model training and subsequently impedes their\neffectiveness. In this work, we propose a novel deep hashing model with only a\nsingle learning objective. Specifically, we show that maximizing the cosine\nsimilarity between the continuous codes and their corresponding binary\northogonal codes can ensure both hash code discriminativeness and quantization\nerror minimization. Further, with this learning objective, code balancing can\nbe achieved by simply using a Batch Normalization (BN) layer and multi-label\nclassification is also straightforward with label smoothing. The result is an\none-loss deep hashing model that removes all the hassles of tuning the weights\nof various losses. Importantly, extensive experiments show that our model is\nhighly effective, outperforming the state-of-the-art multi-loss hashing models\non three large-scale instance retrieval benchmarks, often by significant\nmargins. Code is available at https://github.com/kamwoh/orthohash\n