def torch_calculate_frechet_distance(mu1, sigma1, mu2, sigma2, eps=1e-6):
mu1 = torch.Tensor(mu1.astype("float32"))
sigma1 = torch.Tensor(sigma1.astype("float32"))
mu2 = torch.Tensor(mu2.astype("float32"))
sigma2 = torch.Tensor(sigma2.astype("float32"))
"""Pytorch implementation of the Frechet Distance.
Taken from https://github.com/bioinf-jku/TTUR
The Frechet distance between two multivariate Gaussians X_1 ~ N(mu_1, C_1)
and X_2 ~ N(mu_2, C_2) is
d^2 = ||mu_1 - mu_2||^2 + Tr(C_1 + C_2 - 2*sqrt(C_1*C_2)).
Stable version by Dougal J. Sutherland.
Params:
-- mu1 : Numpy array containing the activations of a layer of the
inception net (like returned by the function 'get_predictions')
for generated samples.
-- mu2 : The sample mean over activations, precalculated on an
representive data set.
-- sigma1: The covariance matrix over activations for generated samples.
-- sigma2: The covariance matrix over activations, precalculated on an
representive data set.
Returns:
-- : The Frechet Distance.
"""
assert mu1.shape == mu2.shape, \
'Training and test mean vectors have different lengths'
assert sigma1.shape == sigma2.shape, \
'Training and test covariances have different dimensions'
diff = mu1 - mu2
# Run 50 itrs of newton-schulz to get the matrix sqrt of sigma1 dot sigma2
covmean = torch.Tensor(
sqrt_newton_schulz(torch.Tensor(sigma1).mm(sigma2).unsqueeze(0),
50)).squeeze()
out = (torch.dot(diff, diff) + torch.trace(sigma1) + torch.trace(sigma2) -
2 * torch.trace(covmean))
return out
def dot(self, x):
return torch.dot(self, x)