This code is implemented for the article "Learning color distributions from bitemporal remote sensing images to update existing building footprints"
The difference between the data distributions of different time-phase images is reduced using the image color translation method
Semantic segmentation model predicts the segmentation results of the images from the latest period
Strictly retain the existing labels of unchanged regions to attain the updated results