PyTorch Implementation of 2D and 3D 'squeeze and excitation' blocks for Fully Convolutional Neural Networks