Update mmdetection to latest github master

Summary:
Wanted to use some of the latest additions to mmdetection; in particular, the UniformAssignment from YOLOF (Free box AP win, even in two stage models!).

I just cloned latest mmdetection and copied into `deeplearning/projects/third_party/mmdetection`.

Reviewed By: ppwwyyxx

Differential Revision: D30598178

fbshipit-source-id: 05fc559534a2fde41b36fc8c088c4e5c1729e7a2

Author: Jon Morton

detectron2/modeling/mmdet_wrapper.py

@@ -45,7 +45,6 @@ def __init__(
         backbone: Union[nn.Module, Mapping],
         neck: Union[nn.Module, Mapping, None] = None,
         *,
-        pretrained_backbone: Optional[str] = None,
         output_shapes: List[ShapeSpec],
         output_names: Optional[List[str]] = None,
     ):
@@ -77,12 +76,11 @@ def __init__(
             neck = build_neck(_to_container(neck))
         self.neck = neck
 
-        # It's confusing that backbone weights are given as a separate argument,
-        # but "neck" weights, if any, are part of neck itself. This is the interface
+        # "Neck" weights, if any, are part of neck itself. This is the interface
         # of mmdet so we follow it. Reference:
         # https://github.com/open-mmlab/mmdetection/blob/master/mmdet/models/detectors/two_stage.py
-        logger.info(f"Initializing mmdet backbone weights: {pretrained_backbone} ...")
-        self.backbone.init_weights(pretrained_backbone)
+        logger.info(f"Initializing mmdet backbone weights...")
+        self.backbone.init_weights()
         # train() in mmdet modules is non-trivial, and has to be explicitly
         # called. Reference:
         # https://github.com/open-mmlab/mmdetection/blob/master/mmdet/models/backbones/resnet.py

tests/modeling/test_mmdet.py

@@ -46,8 +46,6 @@ def test_detector(self):
         MMDetDetector(
             detector=dict(
                 type="MaskRCNN",
-                # skip pretrained model for tests
-                # pretrained="torchvision://resnet50",
                 backbone=dict(
                     type="ResNet",
                     depth=50,
@@ -57,6 +55,8 @@ def test_detector(self):
                     norm_cfg=dict(type="BN", requires_grad=True),
                     norm_eval=True,
                     style="pytorch",
+                    # skip pretrained model for tests
+                    # init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50'))
                 ),
                 neck=dict(
                     type="FPN", in_channels=[256, 512, 1024, 2048], out_channels=256, num_outs=5