基于Mask Scoring R-CNN的高质量数据集快速自动标定方法

Fast Automatic Labeling Method for high quality data sets Based on Mask Scoring R-CNN

摘要:

针对计算机视觉领域人工标定多目标数据集时间冗长的问题，提出一种基于Mask Scoring R-CNN的高质量数据集快速自动标定方法；首先，设计了高质量数据集快速自动标定架构，训练数据自动标定模型并搭建目标分类与标定系统；其次，在对比不同残差网络及引入迁移学习基础上，进一步研究了基于MaskIoU Head的多目标掩膜标定质量评价方法，完成基于Mask Scoring R-CNN的多目标高质量数据集快速自动标定方法设计；最后，以车辆数据为例进行数据集快速自动标定方法验证，实验结果表明，相较于Mask R-CNN和Faster R-CNN方法，Mask Scoring R-CNN方法具有目标数据分类效果好及掩膜分割精度高的优点，检测准确率达到93.4%，且标定速度相较于人工标定速度提升了95.77%。

Abstract:

Aiming at the problem of lengthy manual calibration of multi-target data sets in the field of computer vision,a rapid automatic labeling method based on Mask Scoring R-CNN was proposed. Firstly, a fast automatic labeling framework for high quality data sets is designed. Then, the automatic labeling model of multiple target data is trained and classification and labeling system is built. Secondly, the quality evaluation method of multiple target mask labeling based on MaskIoU Head was further studied based on the comparison of different residual networks and the introduction of transfer learning. Besides,the rapid automatic label method for high quality multiple target data set based on Mask Scoring R-CNN is proposed.Finally, taking vehicle data as an example, the experimental results show that Mask Scoring R-CNN has the advantages of good target data classification effect and high Mask segmentation accuracy compared with the Mask R-CNN and Faster R-CNN. The detection accuracy of the proposed method is 93.4%, and the label speed is 95.77% higher than manual label.