deviandice/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Add CosineEmbeddingLoss to F

Browse Source
This commit is contained in:
James Betker 2020-09-22 17:10:29 -06:00
parent f40beb5460
commit 2e18c4c22d
1 changed files with 6 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

7
codes/models/steps/losses.py
View File

@ -51,6 +51,8 @@ def get_basic_criterion_for_name(name, device):
return nn.L1Loss().to(device) return nn.L1Loss().to(device)
elif name == 'l2': elif name == 'l2':
return nn.MSELoss().to(device) return nn.MSELoss().to(device)
elif name == 'cosine':
return nn.CosineEmbeddingLoss().to(device)
else: else:
raise NotImplementedError raise NotImplementedError
@ -79,7 +81,10 @@ class FeatureLoss(ConfigurableLoss):
with torch.no_grad(): with torch.no_grad():
logits_real = self.netF(state[self.opt['real']]) logits_real = self.netF(state[self.opt['real']])
logits_fake = self.netF(state[self.opt['fake']]) logits_fake = self.netF(state[self.opt['fake']])
return self.criterion(logits_fake, logits_real) if self.opt['criterion'] == 'cosine':
return self.criterion(logits_fake, logits_real, 1)
else:
return self.criterion(logits_fake, logits_real)
# Special form of feature loss which first computes the feature embedding for the truth space, then uses a second # Special form of feature loss which first computes the feature embedding for the truth space, then uses a second