Knowledge distillation (KD) is an effective framework that aims to transfer\nmeaningful information from a large teacher to a smaller student. Generally, KD\noften involves how to define and transfer knowledge. Previous KD methods often\nfocus on mining various forms of knowledge, for example, feature maps and\nrefined information. However, the knowledge is derived from the primary\nsupervised task and thus is highly task-specific. Motivated by the recent\nsuccess of self-supervised representation learning, we propose an auxiliary\nself-supervision augmented task to guide networks to learn more meaningful\nfeatures. Therefore, we can derive soft self-supervision augmented\ndistributions as richer dark knowledge from this task for KD. Unlike previous\nknowledge, this distribution encodes joint knowledge from supervised and\nself-supervised feature learning. Beyond knowledge exploration, we propose to\nappend several auxiliary branches at various hidden layers, to fully take\nadvantage of hierarchical feature maps. Each auxiliary branch is guided to\nlearn self-supervision augmented task and distill this distribution from\nteacher to student. Overall, we call our KD method as Hierarchical\nSelf-Supervision Augmented Knowledge Distillation (HSSAKD). Experiments on\nstandard image classification show that both offline and online HSSAKD achieves\nstate-of-the-art performance in the field of KD. Further transfer experiments\non object detection further verify that HSSAKD can guide the network to learn\nbetter features. The code is available at https://github.com/winycg/HSAKD.\n