Closed felixplum closed 4 years ago
Hi, we did not mean to say that LL is exactly the average pooling. We are saying that it is giving a similar effect (extracting a DC component of an image) to what average pooling does, and it actually is with a constant factor of 1/2 in that the average pooling is 1/2 * [[0.5,0.5], [0.5,0.5]]. The constant factor can be ignored since this operation is combined with the learned weights.
Thanks for the quick answer, that makes sense!
Hi,
in the original paper equivalence between avg. pooling and the LL-component of the decomposition are claimed. However, in your code the L-component evaluates to [[0.5, 0.5], [0.5, 0.5]], which contradicts this statement. Am I missing something?