389 add torchvision resnet50 support

[yiheng-wang-nv]

Fixes #389

This PR is used to enable torchvision's version of resnet50 on PerceptualLoss.
Users can also determine which pretrained weights is used.

[Can-Zhao]

Thank you! LGTM

[Can-Zhao]

@Warvito This PR does not contain breaking change and will not affect existing tutorials. And it adds support for user provided pre-trained weights for perceptual loss. Could it be merged?

[Nic-Ma]

Hi @yiheng-wang-nv @ericspod ,

May I know the status of this PR? Can we merge it now.

Thanks.

[yiheng-wang-nv]

Hi @yiheng-wang-nv @ericspod ,

May I know the status of this PR? Can we merge it now.

Thanks.

It's done. I'm not sure if there are any other requirements to merge a PR in this repo (it's my first contribution), may need @ericspod and @Warvito 's confirm, thanks!

[Nic-Ma]

Hi @ericspod ,

As this PR is necessary for the new MONAI 1.2 bundle and tutorial for Generative AI, I will merge it right now.
If you have any other comments, let's do it in a following PR.

Thanks.

[ericspod]

Hi @Nic-Ma, no problem. I had suggested an enhancement we can follow up with later if there's need.