[fbsync] Adding new ResNet50 weights (#4734)

Summary:
* Update model checkpoint for resnet50.

* Add get_weight method to retrieve weights from name.

* Update the references to support prototype weights.

* Fixing mypy typing.

* Switching to a python 3.6 supported equivalent.

* Add unit-test.

* Add optional num_classes.

Reviewed By: NicolasHug

Differential Revision: D31916330

fbshipit-source-id: 2ac0f9202f62a78078b0917e6730d7fc0925acdf

Author: datumbox

references/classification/train.py

@@ -14,6 +14,12 @@
 from torchvision.transforms.functional import InterpolationMode
 
 
+try:
+    from torchvision.prototype import models as PM
+except ImportError:
+    PM = None
+
+
 def train_one_epoch(model, criterion, optimizer, data_loader, device, epoch, args, model_ema=None, scaler=None):
     model.train()
     metric_logger = utils.MetricLogger(delimiter="  ")
@@ -142,11 +148,18 @@ def load_data(traindir, valdir, args):
         print("Loading dataset_test from {}".format(cache_path))
         dataset_test, _ = torch.load(cache_path)
     else:
+        if not args.weights:
+            preprocessing = presets.ClassificationPresetEval(
+                crop_size=val_crop_size, resize_size=val_resize_size, interpolation=interpolation
+            )
+        else:
+            fn = PM.__dict__[args.model]
+            weights = PM._api.get_weight(fn, args.weights)
+            preprocessing = weights.transforms()
+
         dataset_test = torchvision.datasets.ImageFolder(
             valdir,
-            presets.ClassificationPresetEval(
-                crop_size=val_crop_size, resize_size=val_resize_size, interpolation=interpolation
-            ),
+            preprocessing,
         )
         if args.cache_dataset:
             print("Saving dataset_test to {}".format(cache_path))
@@ -206,7 +219,12 @@ def main(args):
     )
 
     print("Creating model")
-    model = torchvision.models.__dict__[args.model](pretrained=args.pretrained, num_classes=num_classes)
+    if not args.weights:
+        model = torchvision.models.__dict__[args.model](pretrained=args.pretrained, num_classes=num_classes)
+    else:
+        if PM is None:
+            raise ImportError("The prototype module couldn't be found. Please install the latest torchvision nightly.")
+        model = PM.__dict__[args.model](weights=args.weights, num_classes=num_classes)
     model.to(device)
 
     if args.distributed and args.sync_bn:
@@ -455,6 +473,9 @@ def get_args_parser(add_help=True):
         "--train-crop-size", default=224, type=int, help="the random crop size used for training (default: 224)"
     )
 
+    # Prototype models only
+    parser.add_argument("--weights", default=None, type=str, help="the weights enum name to load")
+
     return parser
 
 

test/test_prototype_models.py

@@ -12,6 +12,12 @@ def get_available_classification_models():
     return [k for k, v in models.__dict__.items() if callable(v) and k[0].lower() == k[0] and k[0] != "_"]
 
 
+def test_get_weight():
+    fn = models.resnet50
+    weight_name = "ImageNet1K_RefV2"
+    assert models._api.get_weight(fn, weight_name) == models.ResNet50Weights.ImageNet1K_RefV2
+
+
 @pytest.mark.parametrize("model_name", get_available_classification_models())
 @pytest.mark.parametrize("dev", cpu_and_gpu())
 @pytest.mark.skipif(os.getenv("PYTORCH_TEST_WITH_PROTOTYPE", "0") == "0", reason="Prototype code tests are disabled")

torchvision/prototype/models/_api.py

@@ -1,12 +1,13 @@
 from collections import OrderedDict
 from dataclasses import dataclass, fields
 from enum import Enum
+from inspect import signature
 from typing import Any, Callable, Dict
 
 from ..._internally_replaced_utils import load_state_dict_from_url
 
 
-__all__ = ["Weights", "WeightEntry"]
+__all__ = ["Weights", "WeightEntry", "get_weight"]
 
 
 @dataclass
@@ -74,3 +75,38 @@ def __getattr__(self, name):
             if f.name == name:
                 return object.__getattribute__(self.value, name)
         return super().__getattr__(name)
+
+
+def get_weight(fn: Callable, weight_name: str) -> Weights:
+    """
+    Gets the weight enum of a specific model builder method and weight name combination.
+
+    Args:
+        fn (Callable): The builder method used to create the model.
+        weight_name (str): The name of the weight enum entry of the specific model.
+
+    Returns:
+        Weights: The requested weight enum.
+    """
+    sig = signature(fn)
+    if "weights" not in sig.parameters:
+        raise ValueError("The method is missing the 'weights' argument.")
+
+    ann = signature(fn).parameters["weights"].annotation
+    weights_class = None
+    if isinstance(ann, type) and issubclass(ann, Weights):
+        weights_class = ann
+    else:
+        # handle cases like Union[Optional, T]
+        # TODO: Replace ann.__args__ with typing.get_args(ann) after python >= 3.8
+        for t in ann.__args__:  # type: ignore[union-attr]
+            if isinstance(t, type) and issubclass(t, Weights):
+                weights_class = t
+                break
+
+    if weights_class is None:
+        raise ValueError(
+            "The weight class for the specific method couldn't be retrieved. Make sure the typing info is " "correct."
+        )
+
+    return weights_class.from_str(weight_name)

torchvision/prototype/models/resnet.py

@@ -92,13 +92,13 @@ class ResNet50Weights(Weights):
         },
     )
     ImageNet1K_RefV2 = WeightEntry(
-        url="https://download.pytorch.org/models/resnet50-tmp.pth",
-        transforms=partial(ImageNetEval, crop_size=224),
+        url="https://download.pytorch.org/models/resnet50-f46c3f97.pth",
+        transforms=partial(ImageNetEval, crop_size=224, resize_size=232),
         meta={
             **_common_meta,
             "recipe": "https://github.com/pytorch/vision/issues/3995",
-            "acc@1": 80.352,
-            "acc@5": 95.148,
+            "acc@1": 80.674,
+            "acc@5": 95.166,
         },
     )