/ShallowSeg

Primary LanguagePython

ShallowSeg for Efficient Semantic Segmentation

This repository contains the ShallowSeg model for efficient semantic segmentation.

ShallowSeg follows a ResNet approach with bottleneck modules and makes use of the Concatenated ReLU activation as an efficient means to increase output channel quantity of a convolution, with minimal computation.

Downsampling Blocks

Convolution Blocks

Implementation

This current implementation of ShallowSeg is in TensorFlow ~1.10, and uses TensorFlow Slim layers. TensorFlow Slim is NOT supported in TensorFlow 2.0+.

I am currently aiming to update to a native TF2 implementation.