Implementation of "Few-shot Font Style Transfer with Extraction of Partial Style"
Tested on Windows, but should work on Linux as well.
If you want to use DTS, go to the branch "DTS".
pip install -r requirements.txt
-
Download dataset from Li et al.
-
Unzip it to .\datasets\
-
Start visdom (training visualization):
python -m visdom.server
- To train:
python train.py
NB: Defaults are batch_size = 64, gpu = 0, lambda_SC = 0.
If you get an error ending with:
Pointer addresses:
input: 000000087C570000
output: 00000007854CD800
weight: 0000000785488000
Then decrease the batch size.
- To test:
python test.py
Code derived and rehashed from: