Extending Cutout with a probability of when it is applied

[aisosalo]

How about adding also a probability to vary when Cutout is applied? For example something like this:

class Cutout(object):
    """Randomly mask out one or more patches from an image.
    Args:
        n_holes (int): Number of patches to cut out of each image.
        length (int): The length (in pixels) of each square patch.
        p1 (float): Probability of applying CutOut (default value 1, always applied).
    """
    def __init__(self, n_holes, length, p1=1.0):
        self.n_holes = n_holes
        self.length = length
        self.p1 = p1
		
    def __call__(self, img):
        """
        Args:
            img (Tensor): Tensor image of size (C, H, W).
        Returns:
            Tensor: Image with n_holes of dimension length x length cut out of it.
        """
        if torch.rand(1) > p1:  #np.random.rand(1) > p1:  
            return img
            
            # [...]
            
            return img

So, when np.random.rand(1) is larger than p1 the input image is returned unchanged (most of the time if p1 is small), if I got it right.

[TDeVries]

Yup, this is definitely a possibility!

[gwtaylor]

If you do an experiment, let us know your results!

[aisosalo]

Would you be willing to update your repository with the added functionality in the meantime?