Neural-Style-Transfer-pytorch

Pytorch implementation of Image Style Transfer Using Convolutional Neural Networks. It is my first paper implementation so it would be quite awkward. I recommend you use it only for a reference. The Tutorial Code was very helpful for me to complete my code.

To-do

• Image Style Transfer Using Convolutional Neural Networks (2016) [Paper]
• Combining Markov Random Fields and Convolutional Neural Networks for Image Synthesis (2016) [Paper]
• Perceptual Losses for Real-Time Style Transfer and Super-Resolution (2016) [Paper]
• Fast Patch-based Style Transfer of Arbitrary Style (2016) [Paper]
• A Learned Representation for Artistic Style (2017) [Paper]
• Real-Time Neural Style Transfer for Videos (2017) [Paper]
• Stereoscopic Neural Style Transfer (2018) [Paper]
• Separating Style and Content for Generalized Style Transfer (2018) [Paper]

Usage

To train a model with images you want to merge:

$ python train.py --c_weight=1 \
                  --s_weight=100000 \
                  --content_img='images/dancing.jpg' \
                  --style_img='images/picasso.jpg' \
                  --size=128 --steps=300

To see all training options, run:

$ python train.py --help

which will print:

usage: train.py [-h] [--content_img CONTENT_IMG] [--style_img STYLE_IMG]
                [--size SIZE] [--steps STEPS] [--c_weight C_WEIGHT]
                [--s_weight S_WEIGHT]

optional arguments:
-h, --help                  show this help message and exit
--content_img CONTENT_IMG
--style_img STYLE_IMG
--size SIZE                 if you want to get more clear pictures, increase the
                            size
--steps STEPS
--c_weight C_WEIGHT         weighting factor for content reconstruction
--s_weight S_WEIGHT         weighting factor for style reconstruction

Result

Poster Presentation

Author

Sooyoung Moon / @symoon94