结合上下文注意力的卷积自校正图像语义分割

针对现有图像语义分割中存在小目标对象分割精度不高等问题,提出一种结合上下文注意力的卷积自校正图像语义分割模型.使用上下文注意力机制挖掘局部区域内细粒度特征,结合上下文循环神经网络和残差学习充分挖掘图像的深层隐含语义特征;构建辅助分割模型,在给定图像和边界框注释的情况下生成每像素的标签分布,提出卷积自校正模型,实现分割模...

Convolution self-tuning image semantic segmentation combined with context attention

In the view of the problems of low segmentation accuracy of small objects in existing image semantic segmentation,a convolution self-tuning image semantic segmentation model combined with context attention was proposed.The context attention mechanism was used to mine the fine-grained features in the local region,and the context recurrent neural network and residual learning were combined to fully mine the deep implicit semantic features of the image.An auxiliary segmentation model was constructed to generate the label distribution of each pixel given the image and the bounding box annotation,and a convolution self-tuning model was proposed to realize the dynamic adjustment of the segmentation model.The segmentation accuracy of the proposed model was demonstrated by experiments based on three datasets.The results show that the segmentation accuracy and segmentation effect of the proposed model are significantly higher than other models,and it has good generalization ability.