Random global shift in data transformation/augmentation

I’m trying to reproduce a model described in a paper that I found online, you can found it here.

In section “D. Finetuning the Models” I read:
To encourage the model to be more robust to the exact location of the pleural line we applied a random global shift to the signed distance channel (max ±8%).

I’m not sure what “random global shift to the signed distance channel” means and what kind of transformation I should apply.

Can you help me to understand it please?

I’m using PyTorch in Colab environment.