Attention-Based Network for Semantic Image Segmentation via Adversarial Learning

摘要

As a fundamental research, semantic image segmentation is widely used in the computer vision system. In this paper, we explore the attention mechanism for semantic segmentation to improve the extraction and recovery of information efficiently. Mixed attention modules are designed for the segmentation task, and the attention-based network is the combination by the encoder of Xcep-tion and the decoder of residual connections. Experiments are conducted on the PASCAL VOC dataset, and the proposed method outperforms DccpLabV3Plus. In addition, the adversarial training is deployed based on the attention-based segmentation network, and the experimental results show the performance is further advanced with the addition of adversarial learning.