In case anybody gets confused, there's quite a big editing mistake in this video -- everything from 8:11 to 13:06 is repetition, so just skip to the end of that part.
Great video as always, thanks! Since the inner product is symmetric, aren't we calculating twice the same thing? (k=1,k'=2 and k=2, k'=1 gives the same result, no?)
Yeah..seems like you are right. The gram matrix is computed as the unnormalized covariance matrix (as referred earlier) and a covar matrix is always symmetric. So its transpose should be equal to the actual matrix.
I'm guessing you're still expected to double-sum (k=1,k'=2 and k=2, k'=1). That is why there is a "2" squared in the denominator in the equation at 15:40. Again, just my guess.
Yes, it would be easy to mistake him there. He means to say that whenever one texture is present, the other texture is no more or less present than usual.
In case anybody gets confused, there's quite a big editing mistake in this video -- everything from 8:11 to 13:06 is repetition, so just skip to the end of that part.
I think the equation at 15:02 is missing a square exponent
tru, there's actually a correction about it on coursera
Great video as always, thanks! Since the inner product is symmetric, aren't we calculating twice the same thing? (k=1,k'=2 and k=2, k'=1 gives the same result, no?)
Yeah..seems like you are right. The gram matrix is computed as the unnormalized covariance matrix (as referred earlier) and a covar matrix is always symmetric. So its transpose should be equal to the actual matrix.
I'm guessing you're still expected to double-sum (k=1,k'=2 and k=2, k'=1). That is why there is a "2" squared in the denominator in the equation at 15:40. Again, just my guess.
wonderful
I'm totally lost in this video.
Haha it repeats itself from 8:15, but not entirely
Watch it twice, I don't think it's too difficult.
The two equations (@14:17 & @14:39) seems to be different.
@14:39 there should be a ^2 on the element wise matrix difference
Is there a typo of the Style cost function? Should be the power of the difference between the two style matrices?
what parameters does it train?
There're lots of NG takes :)
No idea what went on in this video, seems a bit too complex
Isn't he defining the non-correlation with the definition of anti-correlation?
Yes, it would be easy to mistake him there. He means to say that whenever one texture is present, the other texture is no more or less present than usual.
Style Cost Function is highly subjective, unless we were to find the ground truth for source images of each painting. PERIOD.