[TorchFix] add visitor for deprecated TorchVision transform (#4615)

This adds a violation when the
[deprecated](https://github.com/pytorch/vision/blob/ddfee23d56700ba84fd28805d5cbdeac2f28f2a7/torchvision/transforms/v2/_deprecated.py#L12-L13)
`torchvision.transforms.v2.ToTensor` transform is imported or used.

For the future, there are two more things to do:

1. The corresponding functional is also
[deprecated](https://github.com/pytorch/vision/blob/ddfee23d56700ba84fd28805d5cbdeac2f28f2a7/torchvision/transforms/v2/functional/_deprecated.py#L10-L11)
and needs the same treatment as the function. If we keep the long
violation message (which is the same message that one would see using
this at runtime), we probably should have a different error code for the
functional to avoid making the message even longer. Otherwise we may
also have both under the same error code.
2. Instead of just complaining about the use of deprecated
functionality, we could also provide a fix. However, this is not
straight forward, since we need to replace `v2.ToTensor()` with
`v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])`.
While that should work everywhere, usually `v2.ToTensor` is not used as
standalone, but rather already as part of a `Compose`, e.g.
    
    ```python
    transform = transforms.Compose([
        transforms.Resize(...),
        transforms.CenterCrop(...),
        transforms.ToTensor(),
        transforms.Normalize(...),
    ])
    ```
    
    or
    
    ```python
    pipeline = [
        transforms.Resize(...),
        transforms.CenterCrop(...),
    ]
    
    if foo:
        pipeline.append(...)
    else:
        pipeline.extend(...)
    
    pipeline.extend([
        transforms.ToTensor(),
        transforms.Normalize(...),
    ])
    transform = transforms.Compose(pipeline)
    ```

In both cases, we want don't want to have an extra `Compose` in there,
but just replace `ToTensor` with `v2.ToImage(),
v2.ToDtype(torch.float32, scale=True)`.

I don't have much experience with `libcst`, but from what I did in this
PR I think 1. is fairly straight forward, while 2. can become a
nightmare.

Author: pmeier

tests/fixtures/vision/checker/to_tensor.py

@@ -0,0 +1,26 @@
+from torchvision.transforms.v2 import ToTensor
+
+from torchvision.transforms.v2 import ToTensor as ToTensorAlias
+
+from torchvision.transforms.v2 import (
+    ToImage,
+    ToTensor,
+    ToDtype,
+)
+
+from torchvision.transforms import v2
+
+v2.ToTensor
+
+from torchvision.transforms import v2 as transforms
+
+transforms.ToTensor
+
+
+import torchvision.transforms.v2 as transforms
+
+transforms.ToTensor
+
+import torchvision.transforms.v2
+
+torchvision.transforms.v2.ToTensor

tests/fixtures/vision/checker/to_tensor.txt

@@ -0,0 +1,7 @@
+1:39 TOR202 The transform `v2.ToTensor()` is deprecated and will be removed in a future release. Instead, please use `v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])`.
+3:39 TOR202 The transform `v2.ToTensor()` is deprecated and will be removed in a future release. Instead, please use `v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])`.
+7:5 TOR202 The transform `v2.ToTensor()` is deprecated and will be removed in a future release. Instead, please use `v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])`.
+13:1 TOR202 The transform `v2.ToTensor()` is deprecated and will be removed in a future release. Instead, please use `v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])`.
+17:1 TOR202 The transform `v2.ToTensor()` is deprecated and will be removed in a future release. Instead, please use `v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])`.
+22:1 TOR202 The transform `v2.ToTensor()` is deprecated and will be removed in a future release. Instead, please use `v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])`.
+26:1 TOR202 The transform `v2.ToTensor()` is deprecated and will be removed in a future release. Instead, please use `v2.Compose([v2.ToImage(), v2.ToDtype(torch.float32, scale=True)])`.

torchfix/torchfix.py

@@ -12,7 +12,9 @@
 
 from .visitors.performance import TorchSynchronizedDataLoaderVisitor
 from .visitors.misc import TorchRequireGradVisitor
-from .visitors.vision import TorchVisionDeprecatedPretrainedVisitor
+from .visitors.vision import (
+    TorchVisionDeprecatedPretrainedVisitor, TorchVisionDeprecatedToTensorVisitor
+)
 
 __version__ = "0.0.3"
 
@@ -27,6 +29,7 @@ def GET_ALL_VISITORS():
         TorchRequireGradVisitor(),
         TorchSynchronizedDataLoaderVisitor(),
         TorchVisionDeprecatedPretrainedVisitor(),
+        TorchVisionDeprecatedToTensorVisitor(),
     ]
 
 

torchfix/visitors/vision/__init__.py

@@ -1,264 +1,2 @@
-from typing import Optional
-
-import libcst as cst
-import libcst.matchers as m
-from libcst.codemod.visitors import ImportItem
-
-from ...common import LintViolation, TorchVisitor
-
-
-class TorchVisionDeprecatedPretrainedVisitor(TorchVisitor):
-    """
-    Find and fix deprecated `pretrained` parameters in TorchVision models.
-
-    Both `pretrained` and `pretrained_backbone` parameters are supported.
-    The parameters are updated to the new `weights` and `weights_backbone` parameters
-    only if the old parameter has explicit literal `True` or `False` value,
-    otherwise only lint violation is emitted.
-    """
-
-    ERROR_CODE = "TOR201"
-
-    # flake8: noqa: E105
-    # fmt: off
-    MODEL_WEIGHTS = {
-        ("mobilenet_v2", "pretrained"): "MobileNet_V2_Weights.IMAGENET1K_V1",
-        ("mobilenet_v3_large", "pretrained"): "MobileNet_V3_Large_Weights.IMAGENET1K_V1",
-        ("mobilenet_v3_small", "pretrained"): "MobileNet_V3_Small_Weights.IMAGENET1K_V1",
-        ("densenet121", "pretrained"): "DenseNet121_Weights.IMAGENET1K_V1",
-        ("densenet161", "pretrained"): "DenseNet161_Weights.IMAGENET1K_V1",
-        ("densenet169", "pretrained"): "DenseNet169_Weights.IMAGENET1K_V1",
-        ("densenet201", "pretrained"): "DenseNet201_Weights.IMAGENET1K_V1",
-        ("detection.maskrcnn_resnet50_fpn", "pretrained"): "MaskRCNN_ResNet50_FPN_Weights.COCO_V1",
-        ("detection.maskrcnn_resnet50_fpn", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("detection.maskrcnn_resnet50_fpn_v2", "pretrained"): "MaskRCNN_ResNet50_FPN_V2_Weights.COCO_V1",
-        ("detection.maskrcnn_resnet50_fpn_v2", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("detection.retinanet_resnet50_fpn", "pretrained"): "RetinaNet_ResNet50_FPN_Weights.COCO_V1",
-        ("detection.retinanet_resnet50_fpn", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("detection.retinanet_resnet50_fpn_v2", "pretrained"): "RetinaNet_ResNet50_FPN_V2_Weights.COCO_V1",
-        ("detection.retinanet_resnet50_fpn_v2", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("optical_flow.raft_large", "pretrained"): "Raft_Large_Weights.C_T_SKHT_V2",
-        ("optical_flow.raft_small", "pretrained"): "Raft_Small_Weights.C_T_V2",
-        ("alexnet", "pretrained"): "AlexNet_Weights.IMAGENET1K_V1",
-        ("convnext_tiny", "pretrained"): "ConvNeXt_Tiny_Weights.IMAGENET1K_V1",
-        ("convnext_small", "pretrained"): "ConvNeXt_Small_Weights.IMAGENET1K_V1",
-        ("convnext_base", "pretrained"): "ConvNeXt_Base_Weights.IMAGENET1K_V1",
-        ("convnext_large", "pretrained"): "ConvNeXt_Large_Weights.IMAGENET1K_V1",
-        ("inception_v3", "pretrained"): "Inception_V3_Weights.IMAGENET1K_V1",
-        ("maxvit_t", "pretrained"): "MaxVit_T_Weights.IMAGENET1K_V1",
-        ("mnasnet0_5", "pretrained"): "MNASNet0_5_Weights.IMAGENET1K_V1",
-        ("mnasnet0_75", "pretrained"): "MNASNet0_75_Weights.IMAGENET1K_V1",
-        ("mnasnet1_0", "pretrained"): "MNASNet1_0_Weights.IMAGENET1K_V1",
-        ("mnasnet1_3", "pretrained"): "MNASNet1_3_Weights.IMAGENET1K_V1",
-        ("detection.fasterrcnn_resnet50_fpn", "pretrained"): "FasterRCNN_ResNet50_FPN_Weights.COCO_V1",
-        ("detection.fasterrcnn_resnet50_fpn", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("detection.fasterrcnn_resnet50_fpn_v2", "pretrained"): "FasterRCNN_ResNet50_FPN_V2_Weights.COCO_V1",
-        ("detection.fasterrcnn_resnet50_fpn_v2", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("detection.fasterrcnn_mobilenet_v3_large_320_fpn", "pretrained"): "FasterRCNN_MobileNet_V3_Large_320_FPN_Weights.COCO_V1",
-        ("detection.fasterrcnn_mobilenet_v3_large_320_fpn", "pretrained_backbone"): "MobileNet_V3_Large_Weights.IMAGENET1K_V1",
-        ("detection.fasterrcnn_mobilenet_v3_large_fpn", "pretrained"): "FasterRCNN_MobileNet_V3_Large_FPN_Weights.COCO_V1",
-        ("detection.fasterrcnn_mobilenet_v3_large_fpn", "pretrained_backbone"): "MobileNet_V3_Large_Weights.IMAGENET1K_V1",
-        ("detection.fcos_resnet50_fpn", "pretrained"): "FCOS_ResNet50_FPN_Weights.COCO_V1",
-        ("detection.fcos_resnet50_fpn", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("segmentation.lraspp_mobilenet_v3_large", "pretrained"): "LRASPP_MobileNet_V3_Large_Weights.COCO_WITH_VOC_LABELS_V1",
-        ("segmentation.lraspp_mobilenet_v3_large", "pretrained_backbone"): "MobileNet_V3_Large_Weights.IMAGENET1K_V1",
-        ("shufflenet_v2_x0_5", "pretrained"): "ShuffleNet_V2_X0_5_Weights.IMAGENET1K_V1",
-        ("shufflenet_v2_x1_0", "pretrained"): "ShuffleNet_V2_X1_0_Weights.IMAGENET1K_V1",
-        ("shufflenet_v2_x1_5", "pretrained"): "ShuffleNet_V2_X1_5_Weights.IMAGENET1K_V1",
-        ("shufflenet_v2_x2_0", "pretrained"): "ShuffleNet_V2_X2_0_Weights.IMAGENET1K_V1",
-        ("squeezenet1_0", "pretrained"): "SqueezeNet1_0_Weights.IMAGENET1K_V1",
-        ("squeezenet1_1", "pretrained"): "SqueezeNet1_1_Weights.IMAGENET1K_V1",
-        ("swin_t", "pretrained"): "Swin_T_Weights.IMAGENET1K_V1",
-        ("swin_s", "pretrained"): "Swin_S_Weights.IMAGENET1K_V1",
-        ("swin_b", "pretrained"): "Swin_B_Weights.IMAGENET1K_V1",
-        ("swin_v2_t", "pretrained"): "Swin_V2_T_Weights.IMAGENET1K_V1",
-        ("swin_v2_s", "pretrained"): "Swin_V2_S_Weights.IMAGENET1K_V1",
-        ("swin_v2_b", "pretrained"): "Swin_V2_B_Weights.IMAGENET1K_V1",
-        ("video.s3d", "pretrained"): "S3D_Weights.KINETICS400_V1",
-        ("video.swin3d_t", "pretrained"): "Swin3D_T_Weights.KINETICS400_V1",
-        ("video.swin3d_s", "pretrained"): "Swin3D_S_Weights.KINETICS400_V1",
-        ("video.swin3d_b", "pretrained"): "Swin3D_B_Weights.KINETICS400_V1",
-        ("vit_b_16", "pretrained"): "ViT_B_16_Weights.IMAGENET1K_V1",
-        ("vit_b_32", "pretrained"): "ViT_B_32_Weights.IMAGENET1K_V1",
-        ("vit_l_16", "pretrained"): "ViT_L_16_Weights.IMAGENET1K_V1",
-        ("vit_l_32", "pretrained"): "ViT_L_32_Weights.IMAGENET1K_V1",
-        ("vit_h_14", "pretrained"): "None",
-        ("vgg11", "pretrained"): "VGG11_Weights.IMAGENET1K_V1",
-        ("vgg11_bn", "pretrained"): "VGG11_BN_Weights.IMAGENET1K_V1",
-        ("vgg13", "pretrained"): "VGG13_Weights.IMAGENET1K_V1",
-        ("vgg13_bn", "pretrained"): "VGG13_BN_Weights.IMAGENET1K_V1",
-        ("vgg16", "pretrained"): "VGG16_Weights.IMAGENET1K_V1",
-        ("vgg16_bn", "pretrained"): "VGG16_BN_Weights.IMAGENET1K_V1",
-        ("vgg19", "pretrained"): "VGG19_Weights.IMAGENET1K_V1",
-        ("vgg19_bn", "pretrained"): "VGG19_BN_Weights.IMAGENET1K_V1",
-        ("video.mvit_v1_b", "pretrained"): "MViT_V1_B_Weights.KINETICS400_V1",
-        ("video.mvit_v2_s", "pretrained"): "MViT_V2_S_Weights.KINETICS400_V1",
-        ("video.r3d_18", "pretrained"): "R3D_18_Weights.KINETICS400_V1",
-        ("video.mc3_18", "pretrained"): "MC3_18_Weights.KINETICS400_V1",
-        ("video.r2plus1d_18", "pretrained"): "R2Plus1D_18_Weights.KINETICS400_V1",
-        ("regnet_y_400mf", "pretrained"): "RegNet_Y_400MF_Weights.IMAGENET1K_V1",
-        ("regnet_y_800mf", "pretrained"): "RegNet_Y_800MF_Weights.IMAGENET1K_V1",
-        ("regnet_y_1_6gf", "pretrained"): "RegNet_Y_1_6GF_Weights.IMAGENET1K_V1",
-        ("regnet_y_3_2gf", "pretrained"): "RegNet_Y_3_2GF_Weights.IMAGENET1K_V1",
-        ("regnet_y_8gf", "pretrained"): "RegNet_Y_8GF_Weights.IMAGENET1K_V1",
-        ("regnet_y_16gf", "pretrained"): "RegNet_Y_16GF_Weights.IMAGENET1K_V1",
-        ("regnet_y_32gf", "pretrained"): "RegNet_Y_32GF_Weights.IMAGENET1K_V1",
-        ("regnet_y_128gf", "pretrained"): "None",
-        ("regnet_x_400mf", "pretrained"): "RegNet_X_400MF_Weights.IMAGENET1K_V1",
-        ("regnet_x_800mf", "pretrained"): "RegNet_X_800MF_Weights.IMAGENET1K_V1",
-        ("regnet_x_1_6gf", "pretrained"): "RegNet_X_1_6GF_Weights.IMAGENET1K_V1",
-        ("regnet_x_3_2gf", "pretrained"): "RegNet_X_3_2GF_Weights.IMAGENET1K_V1",
-        ("regnet_x_8gf", "pretrained"): "RegNet_X_8GF_Weights.IMAGENET1K_V1",
-        ("regnet_x_16gf", "pretrained"): "RegNet_X_16GF_Weights.IMAGENET1K_V1",
-        ("regnet_x_32gf", "pretrained"): "RegNet_X_32GF_Weights.IMAGENET1K_V1",
-        ("resnet18", "pretrained"): "ResNet18_Weights.IMAGENET1K_V1",
-        ("resnet34", "pretrained"): "ResNet34_Weights.IMAGENET1K_V1",
-        ("resnet50", "pretrained"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("resnet101", "pretrained"): "ResNet101_Weights.IMAGENET1K_V1",
-        ("resnet152", "pretrained"): "ResNet152_Weights.IMAGENET1K_V1",
-        ("resnext50_32x4d", "pretrained"): "ResNeXt50_32X4D_Weights.IMAGENET1K_V1",
-        ("resnext101_32x8d", "pretrained"): "ResNeXt101_32X8D_Weights.IMAGENET1K_V1",
-        ("resnext101_64x4d", "pretrained"): "ResNeXt101_64X4D_Weights.IMAGENET1K_V1",
-        ("wide_resnet50_2", "pretrained"): "Wide_ResNet50_2_Weights.IMAGENET1K_V1",
-        ("wide_resnet101_2", "pretrained"): "Wide_ResNet101_2_Weights.IMAGENET1K_V1",
-        ("efficientnet_b0", "pretrained"): "EfficientNet_B0_Weights.IMAGENET1K_V1",
-        ("efficientnet_b1", "pretrained"): "EfficientNet_B1_Weights.IMAGENET1K_V1",
-        ("efficientnet_b2", "pretrained"): "EfficientNet_B2_Weights.IMAGENET1K_V1",
-        ("efficientnet_b3", "pretrained"): "EfficientNet_B3_Weights.IMAGENET1K_V1",
-        ("efficientnet_b4", "pretrained"): "EfficientNet_B4_Weights.IMAGENET1K_V1",
-        ("efficientnet_b5", "pretrained"): "EfficientNet_B5_Weights.IMAGENET1K_V1",
-        ("efficientnet_b6", "pretrained"): "EfficientNet_B6_Weights.IMAGENET1K_V1",
-        ("efficientnet_b7", "pretrained"): "EfficientNet_B7_Weights.IMAGENET1K_V1",
-        ("efficientnet_v2_s", "pretrained"): "EfficientNet_V2_S_Weights.IMAGENET1K_V1",
-        ("efficientnet_v2_m", "pretrained"): "EfficientNet_V2_M_Weights.IMAGENET1K_V1",
-        ("efficientnet_v2_l", "pretrained"): "EfficientNet_V2_L_Weights.IMAGENET1K_V1",
-        ("googlenet", "pretrained"): "GoogLeNet_Weights.IMAGENET1K_V1",
-        ("segmentation.deeplabv3_resnet50", "pretrained"): "DeepLabV3_ResNet50_Weights.COCO_WITH_VOC_LABELS_V1",
-        ("segmentation.deeplabv3_resnet50", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("segmentation.deeplabv3_resnet101", "pretrained"): "DeepLabV3_ResNet101_Weights.COCO_WITH_VOC_LABELS_V1",
-        ("segmentation.deeplabv3_resnet101", "pretrained_backbone"): "ResNet101_Weights.IMAGENET1K_V1",
-        ("segmentation.deeplabv3_mobilenet_v3_large", "pretrained"): "DeepLabV3_MobileNet_V3_Large_Weights.COCO_WITH_VOC_LABELS_V1",
-        ("segmentation.deeplabv3_mobilenet_v3_large", "pretrained_backbone"): "MobileNet_V3_Large_Weights.IMAGENET1K_V1",
-        ("segmentation.fcn_resnet50", "pretrained"): "FCN_ResNet50_Weights.COCO_WITH_VOC_LABELS_V1",
-        ("segmentation.fcn_resnet50", "pretrained_backbone"): "ResNet50_Weights.IMAGENET1K_V1",
-        ("segmentation.fcn_resnet101", "pretrained"): "FCN_ResNet101_Weights.COCO_WITH_VOC_LABELS_V1",
-        ("segmentation.fcn_resnet101", "pretrained_backbone"): "ResNet101_Weights.IMAGENET1K_V1",
-        ("detection.ssd300_vgg16", "pretrained"): "SSD300_VGG16_Weights.COCO_V1",
-        ("detection.ssd300_vgg16", "pretrained_backbone"): "VGG16_Weights.IMAGENET1K_FEATURES",
-        ("detection.ssdlite320_mobilenet_v3_large", "pretrained"): "SSDLite320_MobileNet_V3_Large_Weights.COCO_V1",
-        ("detection.ssdlite320_mobilenet_v3_large", "pretrained_backbone"): "MobileNet_V3_Large_Weights.IMAGENET1K_V1",
-    }
-    # fmt: on
-
-    # The same model can be imported from torchvision.models directly,
-    # or from a submodule like torchvision.models.resnet.
-    MODEL_SUBMODULES = (
-        "alexnet",
-        "convnext",
-        "densenet",
-        "efficientnet",
-        "googlenet",
-        "inception",
-        "mnasnet",
-        "mobilenet",
-        "regnet",
-        "resnet",
-        "shufflenetv2",
-        "squeezenet",
-        "vgg",
-        "vision_transformer",
-        "swin_transformer",
-        "maxvit",
-    )
-
-    def visit_Call(self, node):
-        def _new_arg_and_import(
-            old_arg: cst.Arg, is_backbone: bool
-        ) -> Optional[cst.Arg]:
-            old_arg_name = "pretrained_backbone" if is_backbone else "pretrained"
-            if old_arg is None or (model_name, old_arg_name) not in self.MODEL_WEIGHTS:
-                return None
-            new_arg_name = "weights_backbone" if is_backbone else "weights"
-            weights_arg = None
-            if cst.ensure_type(old_arg.value, cst.Name).value == "True":
-                weights_str = self.MODEL_WEIGHTS[(model_name, old_arg_name)]
-                if is_backbone is False and len(model_name.split(".")) > 1:
-                    # Prepend things like 'detection.' to the weights string
-                    weights_str = model_name.split(".")[0] + "." + weights_str
-                weights_str = "models." + weights_str
-                weights_arg = cst.ensure_type(
-                    cst.parse_expression(f"f({new_arg_name}={weights_str})"), cst.Call
-                ).args[0]
-                self.needed_imports.add(
-                    ImportItem(
-                        module_name="torchvision",
-                        obj_name="models",
-                    )
-                )
-            elif cst.ensure_type(old_arg.value, cst.Name).value == "False":
-                weights_arg = cst.ensure_type(
-                    cst.parse_expression(f"f({new_arg_name}=None)"), cst.Call
-                ).args[0]
-            return weights_arg
-
-        qualified_name = self.get_qualified_name_for_call(node)
-        if qualified_name is None:
-            return
-        if qualified_name.startswith("torchvision.models"):
-            model_name = qualified_name[len("torchvision.models") + 1 :]
-            for submodule in self.MODEL_SUBMODULES:
-                if model_name.startswith(submodule + "."):
-                    model_name = model_name[len(submodule) + 1 :]
-
-            if (model_name, "pretrained") not in self.MODEL_WEIGHTS:
-                return
-
-            message = None
-            pretrained_arg = self.get_specific_arg(node, "pretrained", 0)
-            if pretrained_arg is not None:
-                message = "Parameter `pretrained` is deprecated, please use `weights` instead."
-
-            pretrained_backbone_arg = self.get_specific_arg(
-                node, "pretrained_backbone", 1
-            )
-            if pretrained_backbone_arg is not None:
-                message = "Parameter `pretrained_backbone` is deprecated, please use `weights_backbone` instead."
-
-            replacement_args = list(node.args)
-
-            new_pretrained_arg = _new_arg_and_import(pretrained_arg, is_backbone=False)
-            has_replacement = False
-            if new_pretrained_arg is not None:
-                for pos, arg in enumerate(node.args):
-                    if arg is pretrained_arg:
-                        break
-                replacement_args[pos] = new_pretrained_arg
-                has_replacement = True
-
-            new_pretrained_backbone_arg = _new_arg_and_import(
-                pretrained_backbone_arg, is_backbone=True
-            )
-            if new_pretrained_backbone_arg is not None:
-                for pos, arg in enumerate(node.args):
-                    if arg is pretrained_backbone_arg:
-                        break
-                replacement_args[pos] = new_pretrained_backbone_arg
-                has_replacement = True
-
-            replacement = (
-                node.with_changes(args=replacement_args) if has_replacement else None
-            )
-            if message is not None:
-                position_metadata = self.get_metadata(
-                    cst.metadata.WhitespaceInclusivePositionProvider, node
-                )
-                self.violations.append(
-                    LintViolation(
-                        error_code=self.ERROR_CODE,
-                        message=message,
-                        line=position_metadata.start.line,
-                        column=position_metadata.start.column,
-                        node=node,
-                        replacement=replacement,
-                    )
-                )
+from .pretrained import TorchVisionDeprecatedPretrainedVisitor  # noqa: F401
+from .to_tensor import TorchVisionDeprecatedToTensorVisitor  # noqa: F401