DeepLab
lastmod 2026-11-10created 2020-08-062 edits
DeepLab is a family of semantic segmentation deep learning models published by Google. Its core components are atrous convolution, which widens the receptive field while preserving resolution, and ASPP (Atrous Spatial Pyramid Pooling), which applies atrous convolutions at multiple rates in parallel; v3+ adds an encoder-decoder structure to improve the precision of object boundaries. It is often used as a baseline model for image segmentation tasks that predict a class for each pixel.