群智标注系统中质量管理设计与实现

Design and Implementation of Quality Management in Crowd and Participatory Labeling System

现有的图像标注需要标注人对图像内所有目标进行手工标注,时间漫长,过程复杂,耗时严重,且人具有自主性,在此过程中标注质量存在波动。基于此,论文设计一个通用的智能标注模型,以群智众包的形式将标注任务分派给各个标注者,标注者仅需对智能标注后的图像进行补充优化,大幅度提高了任务的效率,并且通过增加质量管理模块,提高了系统可靠性。通过试验,验证了该模型的可行性和有效性。

The existing image label methods require people to label all targets in the image.This costs too much time and the process is complicated.In addition,people have autonomy influences.All of these factors affect the label results.Based on the above statement,the paper designs a general intelligent label model,which distributes tasks to users in the form of crowdsourcing.Users just need to supplement the intelligent labeled images,which reduces the amount of tasks and improves the efficiency.By adding the quality management module,the tasks'quality has been promoted.At last,the feasibility and effectiveness of the model are veri⁃fied by experiments.