About Channel-Wise Attention(CWA) in the code.

[Hoodythree]

Great work!But I have little question about CWA.In the origianl paper, I see M_i = diag(Mask_i), where diag is putting a vector on the princial diagonal of a diagonal matrix.But in the code below:

foo = [1] * 2 + [0] *  1
bar = []
for i in range(200):
    random.shuffle(foo)
    bar += foo
bar = [bar for i in range(nb_batch)]

I think bar is not a diagonal matrix. Please point out my problem if I misunderstood the operation here.Thanks a lot.

[dongliangchang]

dot([0,1,0].t(), [1,2,3].t()) = [0,2,0].t()
matrix(diag([0,1,0].t()), [1,2,3].t()) = [0,2,0].t()
So, the code and the paper is consistent.
CWA could be visualized as a random channel-dropping operation in each channel group.

[dongliangchang]

[Hoodythree]

Got it.Thanks for the response.

[AND2797]

foo = [1] * 2 + [0] *  1
bar = []
for i in range(200):
    random.shuffle(foo)
    bar += foo
bar = [bar for i in range(nb_batch)]

I had a small question about the code, here foo = [1] * 2 + [0] * 1 is only good for 3 channels correct? If we want to increase the number of channels (say 5), then foo = [1] * 3 + [0] * 2 is necessary, correct?

[dongliangchang]

Yes. Please see Section III.A The Discriminality Component for details.