Squeeze Excite pretrained model

Hello, I’m trying to train the squeeze excite network on imagenet but I forcast it taking weeks to train on my GPU. Is there a way to easily port the weights from another framework like TF or CAFFE into pytorch?

Yeah, your idea is so good. And I thought it before.
You can get the offical response by the pytorch vision github issue.

And the answer is :

I think SENet could be a good addition to torchvision!
But as @alykhantejani mentioned, all models in torchvision should ideally have pre-trained weights trained with pytorch following the same training procedure as in the examples/imagenet, so that it is easily reproducible. The only exception for that for the moment is inception, but we would like to keep those exceptions to a minimum.

I am wondering whether this issue has been solved?

Could anyone share the pytorch-version of the SENet?

I’m not sure if SENet has been trained fully on pyTorch, but I do know you can find equivalently performing models that have been pre-trained, namely the NAS-net family from Google here https://github.com/Cadene/pretrained-models.pytorch