Awesome
Stylebank
Implementation of the paper: StyleBank: An Explicit Representation for Neural Image Style Transfer
Implemented
- Training
- Incremental Learning
Training
- Prepare your content image dataset (I use MS-COCO)
- Prepare your style image dataset
It takes about 1 days to train on a GeForce GTX 1080 Ti. Pre-trained weights are provided. Just load the encoder, decoder and the stylebank you want.
Examples
| Style | Source | Transfered |
|---|---|---|
| <img src="style/00.jpg" width="256"> | <img src="examples/mountain.jpg" width="512"> | <img src="examples/00.jpg" width="512"> |
| <img src="style/01.jpg" width="256"> | <img src="examples/japan.jpg" width="512"> | <img src="examples/01_japan.jpg" width="512"> |
| <img src="style/02.jpg" width="256"> | <img src="examples/bridge.jpg" width="512"> | <img src="examples/02_bridge.jpg" width="512"> |
| <img src="style/02.jpg" width="256"> | <img src="examples/japan101.jpg" width="512"> | <img src="examples/02_japan101.jpg" width="512"> |
| <img src="style/09.jpg" width="256"> | <img src="examples/deer.jpg" width="512"> | <img src="examples/09_deer.jpg" width="512"> |