QuDiDA (QUick and DIrty Domain Adaptation)

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.

Installation

pip install qudida

or

pip install git+https://github.com/arsenyinfo/qudida

Usage

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)

Example

Source image: Target image (style donor): Result with various adaptations: