bilinear sqrt with sign

[happywmzzz]

Hi, it is a concise and useful code for bilinear CNNs, however, from the paper I read about the
" elementwise signed square-root (x ← sign(x)􏰊|x|) and l2 normalization is applied to the matrix A"
which means it should be multiplied by the sign. But in this code just "X = torch.sqrt(X + 1e-5)"

Am I missing something? and even this not same thoroughly, I got the same result (84.2%) which suggests it should be a right answer?

[HaoMood]

Hi, thanks for your comment.

Since X is obtained from the relu5-3 activation of VGG, it is nonnegative elementwise. The bilinear result is nonnegative as well. Therefore, multiplying the sign or not does not affect the result.

Best regards,

Hao

[happywmzzz]

yeah, that's right! I didn't thought the detail numerical value at that layer. thanks again~