How to Reduce Change Detection to Semantic Segmentation

Change detection (CD) aims to identify changes that occur in an image pair taken different times. Prior methods devise specific networks from scratch predict change masks pixel-level, and struggle with general segmentation problems. In this paper, we propose a new paradigm reduces CD semantic which means tailoring existing powerful network solve CD. This conveniently enjoys the mainstream techniques deal problems Hence can concentrate on studying how detect changes. We novel importance insight types exist they should be learned separately. Based it, module named MTF extract information fuse temporal features. high interpretability reveals essential characteristic of And most adapted our module. Finally, C-3PO, at pixel-level. C-3PO achieves state-of-the-art performance without bells whistles. It is simple but effective considered as baseline field. Our code for available https://github.com/DoctorKey/C-3PO.