The code is based on the style-transfer-pytorch, supporting masked image style transfer.
We also deployed the Optimal Transport Loss, and a masked seamless clone-style transfer method.
Also, support using a user word prompt to mask the given image automatically.
Detailed results can be found here:exp/results.ipynb
The main changes happened here: style_transfer/style_transfer.py
*To enable the mask mode and switch to a different loss function, please follow the comments in the codes.
Detailed results can be found here: SAM _generate_mask/predictor_example.ipynb
User Prompt - "sheep"
User Prompt - "bear"
User Prompt - "chair"
-
Style Transfer with Optimal Transport Loss
-
Masked Style Transfer and Mask Strategies
-
Naive Masekd Style Transfer and Mask Strategies
-
MSE-Style Loss, Wasserstein-2 Loss, and Optimal Transport Loss
