Cutout

• @devriesImprovedRegularizationConvolutional2017
• removes constant size square patches randomly by replacing them with any constant value.
• The selection of region is performed by selecting a pixel value randomly and placing a square around it
• Cutout can be expressed as an element-wise multiplication operation $x_{cutout}=x\odot M$, where $x$ is the original image, $M$ is a binary mask of the same size as $x$ with randomly chosen coordinates of a square patch of pixels to be cut out, and $\odot$ denotes element-wise multiplication.