Image Semantic Segmentation using Atrous Pyramid Structures and Fully Convolutional Network
Paper ID : 1121-MVIP2020
Hassan Alikarami *1, MohammadJavad FadaeiEslam2, Farzin Yaghmaee3
1Semnan, Semnan, iran
3semnan, semnan, iran
Understanding an image and understanding the content is one of the most important challenges in machine learning and machine vision. One of the most important ways to address this challenge will be through semantic segmentation. In this paper, we study semantic segmentation using convolution and deconvolution architectures with pyramidal atrous structures. In this method, the basic ResNet101 architecture in feature extraction and later stages of pyramidal structures at the center of the convoluted and reverse cannulated network interfaces are used to combine local information and macro visibility in the feature area. On the other hand, feature extraction layers are transmitted at different levels. The results show 88.1% accuracy and superiority of semantic segmentation over the approaches studied.
semantic segmentation, convolution, deconvolution, pyramidal atrous structures,ResNet101
Status : Paper Accepted (Poster Presentation)