What is DropBlock?
It is a way to perform Regularization. It is almost same as Dropout. However, there is a key difference: DropBlock drops random ‘blocks’ of features to address the disadvantages of Dropout like:
Usage
PyTorch
torchvision.ops.drop_block2d(input, p, block_size, inplace=Fales, eps=1-06, training=True)->Tensor where:
input : The input tensor or 4-dimensions with the first one being its batch.
p : Probability of an element to be dropped.
block_size : Size of the block to drop.
inplace : If set to True, will do this operation in-place.
eps : A value added to the denominator for numerical stability.
training : apply dropblock if is True.
Tensor : The randomly zeroed tensor after dropblock.