Gabriele Cesa

Hi @drewm1980 Unfortunately, I have not much experience with torchscript and torch.jit, so I am not sure how to solve that problem. However, I recommend to check the `.export()` function:...

I see; anyways, if your torch.nn.Module contains some EquivariantModules, you could just manually call the .export() method of the equivariant submodules. For instance if you have a torch.nn.ModuleList which contains...

> Are there any inference time computations or abstractions that you're certain wouldn't just get pruned out by the model optimizer anyway? After setting .eval() mode, my code should be...

Hi @drewm1980 Nice to hear from you again! I think your problem is the einsum in https://github.com/QUVA-Lab/e2cnn/blob/8ce1fbba2dfd2bf12b4ff621d82005c4042a3c0e/e2cnn/nn/modules/r2_conv/basisexpansion_singleblock.py#L99 The one inside GeometricTensor is not usually called inside a neural network. Anyways,...

Hi @namankhetan Thanks for creating this pull request! Unfortunately, I can not merge it yet since it does not pass all the checks (have a look [here](https://github.com/QUVA-Lab/e2cnn/pull/46/checks)). Could you fix...

Hi Kaitlin, That's an interesting idea :) Unfortunately, this is not directly supported for the moment by the library but it would be a very nice feature to include! I...

Hi @blazejdolicki Main questions: 1. If I understood correctly your question, in my code, `inner_type` represents what you called `out_planes`. You can then verify that, when the blocks are instantiated...

Hi @blazejdolicki Sorry for the late reply Regarding the difference between e2wrn.py and e2_wide_resnet.py: the first is a cleaner example I prepared for the tutorial of the library while the...

Hi @blazejdolicki sorry for the late reply. No, the equivariance to 90 deg should be almost perfect, with an absolute error probably lower than 1e-5. At a first look, your...

hi @nijkah Thanks a lot for opening this pull request! Unfortunately, I can not accept it until you sign-off the commit (the DCO check has not passed). Could you sign...