Add detectron2 series of models (#955)
Summary:
Add 10 detectron2 models to the core models collection with preloaded weights file.
```
$ python run.py detectron2_fasterrcnn_r_50_dc5 -d cuda --precision fp16
GPU Time per batch: 25.926 milliseconds
CPU Total Wall Time per batch: 25.926 milliseconds
```
Full list of new models:
- detectron2_fasterrcnn_r_101_c4
- detectron2_fasterrcnn_r_101_dc5
- detectron2_fasterrcnn_r_101_fpn
- detectron2_fasterrcnn_r_50_c4
- detectron2_fasterrcnn_r_50_dc5
- detectron2_fasterrcnn_r_50_fpn
- detectron2_maskrcnn_r_101_c4
- detectron2_maskrcnn_r_101_fpn
- detectron2_maskrcnn_r_50_c4
- detectron2_maskrcnn_r_50_fpn
Pull Request resolved: https://github.com/pytorch/benchmark/pull/955
Reviewed By: terrychenism
Differential Revision: D37193037
Pulled By: xuzhao9
fbshipit-source-id: c16ac85d0944ec098cf4726403b7a558400f9379