guard AuxLogits call with torch.jit.is_scripting()?

[jerryzh168]

Should we guard the check:
https://github.com/pytorch/vision/blob/master/torchvision/models/inception.py#L174
with torch.jit.is_scripting()? since if self.AuxLogits is None we'll be compiling aux = self.AuxLogits(x) and jit will complain that None is used as a function.

I changed it to

       if not torch.jit.is_scripting() and aux_defined:
            aux = self.AuxLogits(x)
        else:
            aux = None

and it worked.

[fmassa]

Hi @jerryzh168

I'm in principle ok with adding this check, but could you give us an example where you faced torchscript errors with inception_v3? We have CI running on it and I've just tried it as well on a simple case and it works fine on torchscript