how to train graph neural networks for different size input for graph classification?

I have a graph dataset at hand (3d mesh) and they have different sizes. I want to use hierarchical pooling but all the literature relies on fixed graph topology so they just use pooling for the one graph since all graphs have the same size. In my case, however, input graphs have different sizes. I want to use the spectral method and it is possible to use that but I have no idea how to handle pooling for each graph one by one.