Mask Scoring R-CNN is a method that improves instance segmentation performance by scoring instance masks more accurately. The method introduces a new network block, called the MaskIoU head, which learns the quality of predicted instance masks by regressing the Intersection-over-Union (IoU) between the predicted mask and its ground truth. This approach calibrates the misalignment between mask quality and mask score, leading to better instance segmentation results. The proposed method, Mask Scoring R-CNN (MS R-CNN), outperforms the state-of-the-art Mask R-CNN on the COCO dataset, achieving consistent and noticeable improvements in performance. The method is simple and effective, and the source code is available for further research and development. The key contributions of this work include the first framework that addresses the problem of scoring instance segmentation hypotheses and the introduction of the MaskIoU head, which is simple and effective. The results show that using mask scores from MS R-CNN instead of only classification confidence leads to a consistent improvement in AP by about 1.5% with various backbone networks. The method is designed for instance segmentation and focuses on scoring the masks, rather than improving instance localization or segmentation mask. The MaskIoU head is implemented within the Mask R-CNN framework and is trained using a simple regression loss. The method is evaluated on the COCO dataset and shows consistent improvements in performance. The results demonstrate that the proposed method provides a new direction for improving instance segmentation.Mask Scoring R-CNN is a method that improves instance segmentation performance by scoring instance masks more accurately. The method introduces a new network block, called the MaskIoU head, which learns the quality of predicted instance masks by regressing the Intersection-over-Union (IoU) between the predicted mask and its ground truth. This approach calibrates the misalignment between mask quality and mask score, leading to better instance segmentation results. The proposed method, Mask Scoring R-CNN (MS R-CNN), outperforms the state-of-the-art Mask R-CNN on the COCO dataset, achieving consistent and noticeable improvements in performance. The method is simple and effective, and the source code is available for further research and development. The key contributions of this work include the first framework that addresses the problem of scoring instance segmentation hypotheses and the introduction of the MaskIoU head, which is simple and effective. The results show that using mask scores from MS R-CNN instead of only classification confidence leads to a consistent improvement in AP by about 1.5% with various backbone networks. The method is designed for instance segmentation and focuses on scoring the masks, rather than improving instance localization or segmentation mask. The MaskIoU head is implemented within the Mask R-CNN framework and is trained using a simple regression loss. The method is evaluated on the COCO dataset and shows consistent improvements in performance. The results demonstrate that the proposed method provides a new direction for improving instance segmentation.