Swin-Transformer icon indicating copy to clipboard operation
Swin-Transformer copied to clipboard

Why to use 3 slices to generate mask?

Open xskxzr opened this issue 3 years ago • 0 comments

Here you use 3 slices to generate mask, which generates 9 different kinds of masks. But if I understand it correctly, only 4 different kinds of masks (as you shown in Figure 4 of your paper, different windows can share the same mask) are needed, thus 2 slices (i.e., slice(0, -self.shift_size) and slice(-self.shift_size, None)) is sufficient, isn't it?

xskxzr avatar Feb 17 '22 06:02 xskxzr