Construction pipeline of the ZTF-NEWm dataset. Both the ZTF-NEWg dataset and the ZTF-NEWr dataset contain 13 000 real sources and 30 000 bogus sources. We selected 6500 real sources from the ZTF-NEWg dataset and 6500 real detections from the ZTF-NEWr dataset. The 13 000 selected samples were combined to form the real source component of the ZTF-NEWm dataset and above them contained at least 4000 first detection sources. Similarly, we randomly chose 15 000 bogus sources from the ZTF-NEWg dataset and 15 000 bogus detections from the ZTF- NEWr dataset. These chosen samples were then combined to create the bogus detection component of the ZTF-NEWm dataset. By combining the real source and bogus detection components, we obtained a dataset (ZTF-NEWm) consisting of 13 000 real sources and 30 000 bogus detections.