In the Mask R-CNN paper the optimizer is described as follows training on MS COCO 2014/2015 dataset for instance segmentation (I believe this is the dataset, correct me if t