QuDiDA is a micro library for very naive though quick pixel level image domain adaptation via scikit-learn transformers.
Is assumed to be used as image augmentation technique, while was not tested in public benchmarks.
pip install qudida
or
pip install git+https://github.com/arsenyinfo/qudida
import cv2
from sklearn.decomposition import PCA
from qudida import DomainAdapter
adapter = DomainAdapter(transformer=PCA(n_components=1), ref_img=cv2.imread('target.png'))
source = cv2.imread('source.png')
result = adapter(source)
cv2.imwrite('../result.png', result)
Source image:
Target image (style donor):
Result with various adaptations:
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent