Pytorch reproduction of the paper "Perceptual Losses for Real Time Style Transfer". Some improvements are made by adopting ideas from "A Learned Representation for Artistic Style"
- Zero-padding is replaced with mirror-padding.
- Transposed convolution is replaced with up-sampling and covolution.
python 3.6.5
pytorch 0.4.1.post2
The residual network is trained with LSUN.
The pre-trained VGG-net provided by pytorch is used.
python main.py
You can run your own experiment by giving parameters manually.
Original content image (Cape Manzamo, Okinawa, Japan):
Style image:
Resulting image:
Samples from LSUN dataset:
| Original Images |  |
 |
 |
|---|---|---|---|
| Resulting Images |  |
 |
 |