FSAU-Net-Tensorflow This is the "FSAU-Net: Convolutional Block Self-Attention" module. The repository also includes an implementation of "Feature Self-Attention Networks" so that you can train and compare between the basic CNN model, the base model of the FSA module and the base models of the SE and ECA modules. The basic CNN models are UNet, ResNet101.
FSA:feature self-attention block FSAU-Net proposes an architectural unit called the Feature Self-Attention (FSA) block that improves representation by using an attention mechanism: making building features dominate and non-building features converge to zero.
Some of the code is now publicly available and we will disclose the code and model after the paper is accepted!