Squeeze and excitation blocks in 3D convnet architectures to forecast physical systems

I am using a temporal 3D U-NET (time dimension + 2 spatial dimensions) to forecast physical features of fluid (pressure, temperature, and velocities) using data from a simulator. I am thinking of using squeeze and excitation in the encoder to capture small scale-large scale movements correlations. So my question is: how can I add the squeeze and excitation block to the 3D U-Net architecture?

Thanks.