5,337
edits
SinGAN: Learning a Generative Model from a Single Natural Image: Difference between revisions
SinGAN: Learning a Generative Model from a Single Natural Image (view source)
Revision as of 14:19, 8 November 2019
, 8 November 2019→Reconstruction Loss
| Line 66: | Line 66: | ||
where the initial noise <math>z^*</math> is drawn once and then fixed during the rest of the training.<br> | where the initial noise <math>z^*</math> is drawn once and then fixed during the rest of the training.<br> | ||
The standard deviation <math>\sigma_n</math> of the noise <math>z_n</math> is proportional to the root mean squared error (RMSE) between the reconstructed patch and the original patch. | The standard deviation <math>\sigma_n</math> of the noise <math>z_n</math> is proportional to the root mean squared error (RMSE) between the reconstructed patch and the original patch. | ||
<syntaxhighlight lang="python"> | |||
loss = nn.MSELoss() | |||
Z_opt = opt.noise_amp*z_opt+z_prev | |||
rec_loss = alpha*loss(netG(Z_opt.detach(),z_prev),real) | |||
rec_loss.backward(retain_graph=True) | |||
</syntaxhighlight> | |||
==Evaluation== | ==Evaluation== | ||