CSCI5210 Project | Pytorch Re-Implementation: Progressive Color Transfer with Dense Semantic Correspondences (SIGGRAPH 2019)
It is just a non-official course project implementation, offcial C++ implementation is here.
export PYTHONPATH=/path/to/the/project/ProgressiveColorTransfer_Pytorch:$PYTHONPATH
cd ProgressiveColorTransfer_Pytorch
bash exps/00-baseline/inference.sh
The patchmatch is superslow since it was a GPU version. You can get intermediate results by reduce the number of feature extraction layers in the configuration file. Some details may different with the original paper, so I just find out a workable setting through experiments.
I did not implement the non-local loss term.
Thanks for the code of Deep Image Analogy PyTorch and Neural Color Transfer PyTorch Implementation.