This text discusses the process of testing transformations before training in computer vision. It highlights the potential pitfalls of blindly boosting dataset size through data augmentation and emphasizes the importance of understanding what transformations are being applied. The post introduces the open-source computer vision libraries FiftyOne and Albumentations, which can be used to generate, visualize, and understand data transformations in real time. It also provides examples of how these tools can be used to test augmentations before including them in a training loop.