- using
l1 loss
. You can see the details inbegan_model.py line 233~236
.
- Maybe, same as in the BEGAN paper. But *different hyper-parameters :).
Name | Global Step 50k | Global Step 100k | Global Step 300k |
---|---|---|---|
BEGAN |
Took about 28800 seconds on GTX 1080.
- Add Loss Function & Explains