DeepLabv3+.pdf

The authors propose DeepLabv3+, an encoder-decoder model for semantic image segmentation, which extends DeepLabv3 by adding a simple yet effective decoder module to refine segmentation results, especially along object boundaries. The model combines the advantages of spatial pyramid pooling and encoder-decoder structures, allowing for the capture of rich contextual information and sharp object boundaries. The authors also explore the use of depthwise separable convolution, adapting the Xception model for semantic segmentation, and apply atrous separable convolution to both the ASPP and decoder modules, resulting in a faster and stronger encoder-decoder network. The proposed model achieves state-of-the-art performance on PASCAL VOC 2012 and Cityscapes datasets, with test set performance of 89.0% and 82.1% without any post-processing.