DeepPhotoStyle_pytorch(中文说明)
Recreating paper "Deep Photo Style Transfer" with pytorch. This project supply semantic segmentation code.
Here are some experiment results
Setup
Install pytorch version 0.4.1 with CUDA Python version: python3.6
git clone
https://github.com/ray075hl/DeepPhotoStyle_pytorch.git
cd DeepPhotoStyle_pytorch
sh download_seg_model.sh
python main.py --style_image path_style_image --content_image path_content_image
Notice
The semantic segmentation result of image pair(style and content) have a huge impact to the quality of transfered image. Check the segmentation result to see whether the relative semantic of image pair as you expected(for example, sky match sky, person match person etc.) or not.
Reference
[1] All the code of semantic segmentation from here Semantic-segmentation-pytorch. I appreciate this fantastic project greatly.
[2] Base framework of neural style transfer. Neural Transfer with PyTorch
[3] Compute laplacian matirx. Closed-form-matting
[4] "Deep Photo Style Transfer"
[5] Post-processing of photo to photo.Visual Attribute Transfer through Deep Image Analogy