The datasets that appear in publications are curated and have been split into training. testing and validation sub-datasets by domain experts. Consequently. machine learning models typically perform well on such split-by-hand prepared datasets. https://goodvibesbadhabiters.shop/product-category/face-mask/