deep-photo-styletransfer

Fork from deep-photo-styletransfer

Modification

• Remove redundancy in examples
• Add -v6 flag in gen_laplacian.m when using octave

Detail Instructions

Installation

• Install Octave

sudo apt-get install octave
sudo apt-get install octave-control octave-image octave-io octave-optim octave-signal octave-statistics

• Install torch

git clone https://github.com/torch/distro.git ~/torch --recursive
cd ~/torch; 
bash install-deps; # maybe skip 
./install.sh
sudo apt-get install libmatio2
luarocks install matio
luarocks install loadcaffe

• Download pre-trained VGG-19 model

sh models/download_models.sh

• Compile extra cuda lib (Modify the first two lines in makefile, then make it)

make clean && make

Running

• Please make sure the all png images has longer side less than 700px, resize_images.py will help to do that.

• Generate laplacian matrix (Takes really long, matlab is faster)

cd gen_laplacian;
octave gen_laplacian.m --no-gui

• Masks with the same color is in the same group.
• Ok let's go

python gen_all.py

• Test successfully on TITAN X with 12GB memory, no OOM issues.