To avoid any bias, we need to create two separate and independent sets of images. Indeed, you don't want to evaluate the performance of a model on the same set of images it has been trained on. Hence, we distinguish between training and validation sets. By default, when you do not specify any information, all the images that you add to a project are added to the training set. To create or change the validation set, you need to click on Perform a validation split on the navigation bar in the Galleries section.