Update training references from legacy models

references/classification/README.md

@@ -42,20 +42,20 @@ torchrun --nproc_per_node=8 train.py --model inception_v3\
       --val-resize-size 342 --val-crop-size 299 --train-crop-size 299 --test-only --pretrained
 ```
 
-### ResNext-50 32x4d
+### ResNet
 ```
-torchrun --nproc_per_node=8 train.py\
-    --model resnext50_32x4d --epochs 100
+torchrun --nproc_per_node=8 train.py --model $MODEL
 ```
 
+Here `$MODEL` is one of `resnet18`, `resnet34`, `resnet50`, `resnet101` or `resnet152`.
 
-### ResNext-101 32x8d
-
+### ResNext
 ```
 torchrun --nproc_per_node=8 train.py\
-    --model resnext101_32x8d --epochs 100
+    --model $MODEL --epochs 100
 ```
 
+Here `$MODEL` is one of `resnext50_32x4d` or `resnext101_32x8d`.
 Note that the above command corresponds to a single node with 8 GPUs. If you use
 a different number of GPUs and/or a different batch size, then the learning rate
 should be scaled accordingly. For example, the pretrained model provided by

torchvision/prototype/models/alexnet.py

@@ -13,15 +13,14 @@
 __all__ = ["AlexNet", "AlexNetWeights", "alexnet"]
 
 
-_common_meta = {"size": (224, 224), "categories": _IMAGENET_CATEGORIES, "interpolation": InterpolationMode.BILINEAR}
-
-
 class AlexNetWeights(Weights):
     ImageNet1K_RefV1 = WeightEntry(
         url="https://download.pytorch.org/models/alexnet-owt-7be5be79.pth",
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
-            **_common_meta,
+            "size": (224, 224),
+            "categories": _IMAGENET_CATEGORIES,
+            "interpolation": InterpolationMode.BILINEAR,
             "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#alexnet-and-vgg",
             "acc@1": 56.522,
             "acc@5": 79.066,

torchvision/prototype/models/densenet.py

@@ -63,55 +63,56 @@ def _densenet(
     return model
 
 
-_common_meta = {"size": (224, 224), "categories": _IMAGENET_CATEGORIES, "interpolation": InterpolationMode.BILINEAR}
+_common_meta = {
+    "size": (224, 224),
+    "categories": _IMAGENET_CATEGORIES,
+    "interpolation": InterpolationMode.BILINEAR,
+    "recipe": None,  # weights ported from LuaTorch
+}
 
 
 class DenseNet121Weights(Weights):
-    ImageNet1K_RefV1 = WeightEntry(
+    ImageNet1K_Community = WeightEntry(
         url="https://download.pytorch.org/models/densenet121-a639ec97.pth",
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_common_meta,
-            "recipe": "",
             "acc@1": 74.434,
             "acc@5": 91.972,
         },
     )
 
 
 class DenseNet161Weights(Weights):
-    ImageNet1K_RefV1 = WeightEntry(
+    ImageNet1K_Community = WeightEntry(
         url="https://download.pytorch.org/models/densenet161-8d451a50.pth",
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_common_meta,
-            "recipe": "",
             "acc@1": 77.138,
             "acc@5": 93.560,
         },
     )
 
 
 class DenseNet169Weights(Weights):
-    ImageNet1K_RefV1 = WeightEntry(
+    ImageNet1K_Community = WeightEntry(
         url="https://download.pytorch.org/models/densenet169-b2777c0a.pth",
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_common_meta,
-            "recipe": "",
             "acc@1": 75.600,
             "acc@5": 92.806,
         },
     )
 
 
 class DenseNet201Weights(Weights):
-    ImageNet1K_RefV1 = WeightEntry(
+    ImageNet1K_Community = WeightEntry(
         url="https://download.pytorch.org/models/densenet201-c1103571.pth",
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_common_meta,
-            "recipe": "",
             "acc@1": 76.896,
             "acc@5": 93.370,
         },
@@ -121,7 +122,7 @@ class DenseNet201Weights(Weights):
 def densenet121(weights: Optional[DenseNet121Weights] = None, progress: bool = True, **kwargs: Any) -> DenseNet:
     if "pretrained" in kwargs:
         warnings.warn("The argument pretrained is deprecated, please use weights instead.")
-        weights = DenseNet121Weights.ImageNet1K_RefV1 if kwargs.pop("pretrained") else None
+        weights = DenseNet121Weights.ImageNet1K_Community if kwargs.pop("pretrained") else None
     weights = DenseNet121Weights.verify(weights)
 
     return _densenet(32, (6, 12, 24, 16), 64, weights, progress, **kwargs)
@@ -130,7 +131,7 @@ def densenet121(weights: Optional[DenseNet121Weights] = None, progress: bool = T
 def densenet161(weights: Optional[DenseNet161Weights] = None, progress: bool = True, **kwargs: Any) -> DenseNet:
     if "pretrained" in kwargs:
         warnings.warn("The argument pretrained is deprecated, please use weights instead.")
-        weights = DenseNet161Weights.ImageNet1K_RefV1 if kwargs.pop("pretrained") else None
+        weights = DenseNet161Weights.ImageNet1K_Community if kwargs.pop("pretrained") else None
     weights = DenseNet161Weights.verify(weights)
 
     return _densenet(48, (6, 12, 36, 24), 96, weights, progress, **kwargs)
@@ -139,7 +140,7 @@ def densenet161(weights: Optional[DenseNet161Weights] = None, progress: bool = T
 def densenet169(weights: Optional[DenseNet169Weights] = None, progress: bool = True, **kwargs: Any) -> DenseNet:
     if "pretrained" in kwargs:
         warnings.warn("The argument pretrained is deprecated, please use weights instead.")
-        weights = DenseNet169Weights.ImageNet1K_RefV1 if kwargs.pop("pretrained") else None
+        weights = DenseNet169Weights.ImageNet1K_Community if kwargs.pop("pretrained") else None
     weights = DenseNet169Weights.verify(weights)
 
     return _densenet(32, (6, 12, 32, 32), 64, weights, progress, **kwargs)
@@ -148,7 +149,7 @@ def densenet169(weights: Optional[DenseNet169Weights] = None, progress: bool = T
 def densenet201(weights: Optional[DenseNet201Weights] = None, progress: bool = True, **kwargs: Any) -> DenseNet:
     if "pretrained" in kwargs:
         warnings.warn("The argument pretrained is deprecated, please use weights instead.")
-        weights = DenseNet201Weights.ImageNet1K_RefV1 if kwargs.pop("pretrained") else None
+        weights = DenseNet201Weights.ImageNet1K_Community if kwargs.pop("pretrained") else None
     weights = DenseNet201Weights.verify(weights)
 
     return _densenet(32, (6, 12, 48, 32), 64, weights, progress, **kwargs)

torchvision/prototype/models/efficientnet.py

@@ -62,7 +62,11 @@ def _efficientnet(
     return model
 
 
-_common_meta = {"categories": _IMAGENET_CATEGORIES, "interpolation": InterpolationMode.BICUBIC}
+_common_meta = {
+    "categories": _IMAGENET_CATEGORIES,
+    "interpolation": InterpolationMode.BICUBIC,
+    "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
+}
 
 
 class EfficientNetB0Weights(Weights):
@@ -72,7 +76,6 @@ class EfficientNetB0Weights(Weights):
         meta={
             **_common_meta,
             "size": (224, 224),
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
             "acc@1": 77.692,
             "acc@5": 93.532,
         },
@@ -86,7 +89,6 @@ class EfficientNetB1Weights(Weights):
         meta={
             **_common_meta,
             "size": (240, 240),
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
             "acc@1": 78.642,
             "acc@5": 94.186,
         },
@@ -100,7 +102,6 @@ class EfficientNetB2Weights(Weights):
         meta={
             **_common_meta,
             "size": (288, 288),
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
             "acc@1": 80.608,
             "acc@5": 95.310,
         },
@@ -114,7 +115,6 @@ class EfficientNetB3Weights(Weights):
         meta={
             **_common_meta,
             "size": (300, 300),
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
             "acc@1": 82.008,
             "acc@5": 96.054,
         },
@@ -128,7 +128,6 @@ class EfficientNetB4Weights(Weights):
         meta={
             **_common_meta,
             "size": (380, 380),
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
             "acc@1": 83.384,
             "acc@5": 96.594,
         },
@@ -142,7 +141,6 @@ class EfficientNetB5Weights(Weights):
         meta={
             **_common_meta,
             "size": (456, 456),
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
             "acc@1": 83.444,
             "acc@5": 96.628,
         },
@@ -156,7 +154,6 @@ class EfficientNetB6Weights(Weights):
         meta={
             **_common_meta,
             "size": (528, 528),
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
             "acc@1": 84.008,
             "acc@5": 96.916,
         },
@@ -170,7 +167,6 @@ class EfficientNetB7Weights(Weights):
         meta={
             **_common_meta,
             "size": (600, 600),
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#efficientnet",
             "acc@1": 84.122,
             "acc@5": 96.908,
         },

torchvision/prototype/models/googlenet.py

@@ -13,15 +13,14 @@
 __all__ = ["GoogLeNet", "GoogLeNetOutputs", "_GoogLeNetOutputs", "GoogLeNetWeights", "googlenet"]
 
 
-_common_meta = {"size": (224, 224), "categories": _IMAGENET_CATEGORIES, "interpolation": InterpolationMode.BILINEAR}
-
-
 class GoogLeNetWeights(Weights):
     ImageNet1K_Community = WeightEntry(
         url="https://download.pytorch.org/models/googlenet-1378be20.pth",
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
-            **_common_meta,
+            "size": (224, 224),
+            "categories": _IMAGENET_CATEGORIES,
+            "interpolation": InterpolationMode.BILINEAR,
             "recipe": "https://github.com/TheCodez/examples/blob/inception/imagenet/README.md#googlenet",
             "acc@1": 69.778,
             "acc@5": 89.530,

torchvision/prototype/models/inception.py

@@ -13,15 +13,14 @@
 __all__ = ["Inception3", "InceptionOutputs", "_InceptionOutputs", "Inception3Weights", "inception_v3"]
 
 
-_common_meta = {"size": (299, 299), "categories": _IMAGENET_CATEGORIES, "interpolation": InterpolationMode.BILINEAR}
-
-
 class Inception3Weights(Weights):
     ImageNet1K_TFV1 = WeightEntry(
         url="https://download.pytorch.org/models/inception_v3_google-0cc3c7bd.pth",
         transforms=partial(ImageNetEval, crop_size=299, resize_size=342),
         meta={
-            **_common_meta,
+            "size": (299, 299),
+            "categories": _IMAGENET_CATEGORIES,
+            "interpolation": InterpolationMode.BILINEAR,
             "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#inception-v3",
             "acc@1": 77.294,
             "acc@5": 93.450,

torchvision/prototype/models/mnasnet.py

@@ -23,7 +23,12 @@
 ]
 
 
-_common_meta = {"size": (224, 224), "categories": _IMAGENET_CATEGORIES, "interpolation": InterpolationMode.BILINEAR}
+_common_meta = {
+    "size": (224, 224),
+    "categories": _IMAGENET_CATEGORIES,
+    "interpolation": InterpolationMode.BILINEAR,
+    "recipe": "https://github.com/1e100/mnasnet_trainer",
+}
 
 
 class MNASNet0_5Weights(Weights):
@@ -32,7 +37,6 @@ class MNASNet0_5Weights(Weights):
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_common_meta,
-            "recipe": "https://github.com/1e100/mnasnet_trainer",
             "acc@1": 67.734,
             "acc@5": 87.490,
         },
@@ -50,7 +54,6 @@ class MNASNet1_0Weights(Weights):
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_common_meta,
-            "recipe": "https://github.com/1e100/mnasnet_trainer",
             "acc@1": 73.456,
             "acc@5": 91.510,
         },

torchvision/prototype/models/mobilenetv2.py

@@ -13,15 +13,14 @@
 __all__ = ["MobileNetV2", "MobileNetV2Weights", "mobilenet_v2"]
 
 
-_common_meta = {"size": (224, 224), "categories": _IMAGENET_CATEGORIES, "interpolation": InterpolationMode.BILINEAR}
-
-
 class MobileNetV2Weights(Weights):
     ImageNet1K_RefV1 = WeightEntry(
         url="https://download.pytorch.org/models/mobilenet_v2-b0353104.pth",
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
-            **_common_meta,
+            "size": (224, 224),
+            "categories": _IMAGENET_CATEGORIES,
+            "interpolation": InterpolationMode.BILINEAR,
             "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#mobilenetv2",
             "acc@1": 71.878,
             "acc@5": 90.286,

torchvision/prototype/models/mobilenetv3.py

@@ -37,7 +37,12 @@ def _mobilenet_v3(
     return model
 
 
-_common_meta = {"size": (224, 224), "categories": _IMAGENET_CATEGORIES, "interpolation": InterpolationMode.BILINEAR}
+_common_meta = {
+    "size": (224, 224),
+    "categories": _IMAGENET_CATEGORIES,
+    "interpolation": InterpolationMode.BILINEAR,
+    "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#mobilenetv3-large--small",
+}
 
 
 class MobileNetV3LargeWeights(Weights):
@@ -46,7 +51,6 @@ class MobileNetV3LargeWeights(Weights):
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_common_meta,
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#mobilenetv3-large--small",
             "acc@1": 74.042,
             "acc@5": 91.340,
         },
@@ -59,7 +63,6 @@ class MobileNetV3SmallWeights(Weights):
         transforms=partial(ImageNetEval, crop_size=224),
         meta={
             **_common_meta,
-            "recipe": "https://github.com/pytorch/vision/tree/main/references/classification#mobilenetv3-large--small",
             "acc@1": 67.668,
             "acc@5": 87.402,
         },