目的 人机交互信息在交互式图像分割过程中具有重要意义，为了提高交互信息的使用效率，文中提出一种优化方法。方法 提出一种非对称注意力结构，将交互信息通过该结构融合到交互式图像分割算法(IOG)的特征提取网络中。该算法能够进一步强化关键点信息对图像分割所起到的引导作用。结果 非对称注意力结构能够在不增加交互成本的条件下，在PASCAL数据集上达到92.2%的准确率，比目前最好的IOG分割算法提高了0.2%。仅在小样本PASCAL数据集上训练时，文中算法具有更明显的优势，比现有最好的IOG算法的准确率提高了1.3%。结论 通过中文的非对称注意力结构，能够在不增加交互成本的同时提升网络的分割精度。

In the process of interactive image segmentation, human-computer interaction plays an important role. For higher efficiency of human-computer interaction, this paper describes a structure of asymmetric key points attention, which can integrate human-computer interaction into the feature extraction network of interactive object segmentation with inside-outside guidance (IOG), based on guidance reinforcement of IOG for image segmentation of key points. This structure enhanced the accuracy to 92.2% without increasing the cost of interaction on PASCAL, 0.2% higher IOG (current best segmentation algorithm). While only training on PASCAL, the accuracy of this structure was obviously 1.3% higher than IOG. Under the assistance of the structure of asymmetric key points attention, the accuracy of segmentation can be improved without increasing the cost of interaction.