Runs MaskR-CNN (Detectron) on residual images of a source separation algorithm to identify undetected sources.

The goal here is to identify previously undetected galaxies from the model of a given scene and it's residual. We run a detection algorithm which is a modified version of an implementation of the Mask R-CNN architecture. The project webpage has a detailed explanation about the nature of the problem, our analysis method and results. Since we only want the position of undetected sources we don't generate segmentation maps for the detections.