jaehong31/CGES

Figure2(c) competition among different filters for the same spatial feature

lizhenstat opened this issue · 0 comments

Hi, I have a question on Figure 2(c)
Exclusive sparsity enforces each convolution filter to learn features that are as different as possible, by promoting competition among the filters for the same spatial feature.
Since convolution kernels is shared among different positions in a feature map,
what do you mean by the same spatial feature here

while in the following graph, the exclusive lasso sparsity is defined as follows:
while defining each group g as the same feature across multiple convolution filters

these two definations does not seem to mean the same thing
thanks a lot