Corrected value for adversarial loss. Don't refactor math the day aft…

…er stopping coffee.
alexjc · Oct 31, 2016 · 02d2fca · 02d2fca
1 parent 0c9937a
commit 02d2fca
Show file tree

Hide file tree

Showing 2 changed files with 2 additions and 2 deletions.
diff --git a/README.rst b/README.rst
@@ -64,7 +64,7 @@ Pre-trained models are provided in the GitHub releases.  Training your own is a
     
     # Train the model using an adversarial setup based on [4] below.
     python3.4 enhance.py --train "data/*.jpg" --model custom --scales=2 --epochs=250 \
-             --perceptual-layer=conv5_2 --smoothness-weight=2e4 --adversary-weight=2e5 \
+             --perceptual-layer=conv5_2 --smoothness-weight=2e4 --adversary-weight=1e3 \
              --generator-start=5 --discriminator-start=0 --adversarial-start=5 \
              --discriminator-size=64
 

diff --git a/enhance.py b/enhance.py
@@ -374,7 +374,7 @@ def loss_total_variation(self, x):
         return T.mean(((x[:,:,:-1,:-1] - x[:,:,1:,:-1])**2 + (x[:,:,:-1,:-1] - x[:,:,:-1,1:])**2)**1.25)
 
     def loss_adversarial(self, d):
-        return T.mean(1.0 - T.nnet.softplus(d[args.batch_size:]))
+        return T.mean(1.0 - T.nnet.softminus(d[args.batch_size:]))
 
     def loss_discriminator(self, d):
         return T.mean(T.nnet.softminus(d[args.batch_size:]) - T.nnet.softplus(d[:args.batch_size]))