[fbsync] Add new weights for wide_resnet101_2 model (#4906)

Summary: * Add new weights for wide_resnet101_2 model * Updated accuracies Reviewed By: datumbox Differential Revision: D32470481 fbshipit-source-id: 145391d20ca8c6da3cee44239a1df93eb350d733 Co-authored-by: Vasilis Vryniotis <[email protected]>
pytorch · Nov 17, 2021 · 20b98d6 · 20b98d6
1 parent fe00a8d
commit 20b98d6
Showing 1 changed file with 10 additions and 0 deletions.
diff --git a/torchvision/prototype/models/resnet.py b/torchvision/prototype/models/resnet.py
@@ -219,6 +219,16 @@ class WideResNet101_2Weights(Weights):
             "acc@5": 94.284,
         },
     )
+    ImageNet1K_RefV2 = WeightEntry(
+        url="https://download.pytorch.org/models/wide_resnet101_2-b8680a8c.pth",
+        transforms=partial(ImageNetEval, crop_size=224, resize_size=232),
+        meta={
+            **_common_meta,
+            "recipe": "https://github.com/pytorch/vision/issues/3995",
+            "acc@1": 82.492,
+            "acc@5": 96.110,
+        },
+    )
 
 
 def resnet18(weights: Optional[ResNet18Weights] = None, progress: bool = True, **kwargs: Any) -> ResNet: