add the pretrained resnext101_64x4d model

[tangjiuqi097]

🚀 Feature

add the pretrained resnext101_64x4d model

Motivation

In object detection, the resnext101_64x4d is a widely used backbone.
But the torchvision model only provide the resnext101_32x8d pretrained models.
If we want to use the resnext101_64x4d in object detection, it cannot be smoothly to change.

[oke-aditya]

Hello. @tangjiuqi097 . The discussion for adding new models is in #2707. Feel free to comment there.
I will add this to the list.