性能超越何恺明Mask R-CNN！华科大开源图像分割新方法

CVPR是IEEE Conference on Computer Vision and Pattern Recongnition的缩写，即IEEE国际计算机视觉与模式识别会议。该会议是由IEEE举办的计算机视觉和模式识别领域的顶级会议。

CVPR 2019一共收到5165篇有效投递，一共接收了1300篇。本文选取了其中的口头报告论文进行推荐。

• 论文题目

  Mask Scoring R-CNN

• 作者

  Zhaojin Huang, Lichao Huang, Yongchao Gong, Chang Huang, Xinggang Wang

• 会议/年份

  CVPR 2019

• 链接

  https://arxiv.org/abs/1903.00241v1

• Abstract

  Letting a deep network be aware of the quality of its own predictions is an interesting yet important problem. In the task of instance segmentation, the confidence of instance classification is used as mask quality score in most instance segmentation frameworks. However, the mask quality, quantified as the IoU between the instance mask and its ground truth, is usually not well correlated with classification score. In this paper, we study this problem and propose Mask Scoring R-CNN which contains a network block to learn the quality of the predicted instance masks. The proposed network block takes the instance feature and the corresponding predicted mask together to regress the mask IoU. The mask scoring strategy calibrates the misalignment between mask quality and mask score, and improves instance segmentation performance by prioritizing more accurate mask predictions during COCO AP evaluation. By extensive evaluations on the COCO dataset, Mask Scoring R-CNN brings consistent and noticeable gain with different models, and outperforms the state-of-the-art Mask R-CNN. We hope our simple and effective approach will provide a new direction for improving instance segmentation. The source code of our method is available at \url{this https URL}.

推荐理由

华中科技大学的黄钊金作为一作完成的研究Mask Scoring R-CNN，在COCO图像实例分割任务上超越了何恺明的Mask R-CNN，拿下了计算机视觉顶会CVPR 2019的口头报告，也就是说这篇论文从5000多篇投稿中脱颖而出，成为最顶尖的5.6%。

这篇论文中，研究人员提出了一种给算法的“实例分割假设”打分的新方法。这个分数打得是否准确，就会影响实例分割模型的性能。而Mask R-CNN等前辈，用的打分方法就不太合适。这些模型在实例分割任务里，虽然输出结果是一个蒙版，但打分却是和边界框目标检测共享的，都是针对目标区域分类置信度算出来的分数。这个分数，和图像分割蒙版的质量可未必一致，用来评价蒙版的质量，可能就会出偏差。

于是，这篇CPR 2019论文就提出了一种新的打分方法：给蒙版打分，他们称之为蒙版得分（mask score）。

上图为COCO 2017测试集（Test-De set）上MS R-CNN和其他实例分割方法的成绩对比。无论基干网络是纯粹的ResNet-101，还是用了DCN、FPN，MS R-CNN的AP成绩都比Mask R-CNN高出一点几个百分点。

传送门：

论文地址：

https://arxiv.org/pdf/1903.00241v1.pdf

该项目已开源：

https://github.com/zjhuang22/maskscoring_rcnn