[BC-breaking] Update KeypointRCNN weights (#1609)

* Update KeypointRCNN weights with correct file

* Fix model

* Fix

torchvision/models/detection/keypoint_rcnn.py

@@ -259,8 +259,11 @@ class KeypointRCNNPredictor(nn.Module):
 
 
 model_urls = {
-    'keypointrcnn_resnet50_fpn_coco':
+    # legacy model for BC reasons, see https://github.com/pytorch/vision/issues/1606
+    'keypointrcnn_resnet50_fpn_coco_legacy':
         'https://download.pytorch.org/models/keypointrcnn_resnet50_fpn_coco-9f466800.pth',
+    'keypointrcnn_resnet50_fpn_coco':
+        'https://download.pytorch.org/models/keypointrcnn_resnet50_fpn_coco-fc266e95.pth',
 }
 
 
@@ -312,7 +315,10 @@ def keypointrcnn_resnet50_fpn(pretrained=False, progress=True,
     backbone = resnet_fpn_backbone('resnet50', pretrained_backbone)
     model = KeypointRCNN(backbone, num_classes, num_keypoints=num_keypoints, **kwargs)
     if pretrained:
-        state_dict = load_state_dict_from_url(model_urls['keypointrcnn_resnet50_fpn_coco'],
+        key = 'keypointrcnn_resnet50_fpn_coco'
+        if pretrained == 'legacy':
+            key += '_legacy'
+        state_dict = load_state_dict_from_url(model_urls[key],
                                               progress=progress)
         model.load_state_dict(state_dict)
     return model