Project from a group conv. layer to a subgroup conv. layer

[ph-code-repo]

I was watching the Group Equivariant Deep Learning lectures by Erik Bekkers, and they mentioned projecting from 3D group convolutional layers to a 2D feature map using eg. max pooling. ESCNN seems to have some pooling functionality, but treats all channels independently.
What is the best way to pool layers of a larger group, say the dihedral group, into layers of a subgroup, like the cyclic group? In this situation, I might want my detections to be rotation equivariant, but reflection invariant, for example.
From further reading, it looks a lot like coset pooling as described in these slides. Is there a practical implementation of this in the library?

Thanks.