Most deep learning tasks these days are best solved by using pretrained models or established architectures. Training, data gathering, and preprocessing have the highest chance of success when we use best practices, stick to good literature and papers, and avoid getting too creative. After all, they are called “best” practices for a reason. However, every so often, there are useful little hacks and tricks that either never made it into papers or are buried deep in the gigantic mountain that is arXiv. In this session, I will present a number of strange little tidbits from our everyday work that helped us out at one point. These techniques are either very obscure, defy common wisdom, or are actually best practices that are often ignored, even by seasoned professionals. There will be no common theme or thread other than that every technique is either weird, unusual, little-known, or fun. Preferably all of the above, much like a renaissance chamber of curiosities or “Wunderkammer”.