Add missing `handle_legacy_interface()` calls (#6565)

* Add `handle_legacy_interface()` to all new models.

* Fix imports

* Addressing review comments.

* Fix linter

* Addressing further comments.

test/test_extended_models.py

@@ -335,3 +335,17 @@ class TestHandleLegacyInterface:
 
         with pytest.raises(ValueError, match="weights"):
             builder(pretrained=True, flag=False)
+
+    @pytest.mark.parametrize(
+        "model_fn",
+        [fn for fn in TM.list_model_fns(models) if fn.__name__ not in {"vit_h_14", "regnet_y_128gf"}]
+        + TM.list_model_fns(models.detection)
+        + TM.list_model_fns(models.quantization)
+        + TM.list_model_fns(models.segmentation)
+        + TM.list_model_fns(models.video)
+        + TM.list_model_fns(models.optical_flow),
+    )
+    @run_if_test_with_extended
+    def test_pretrained_deprecation(self, model_fn):
+        with pytest.warns(UserWarning, match="deprecated"):
+            model_fn(pretrained=True)

torchvision/models/detection/faster_rcnn.py

@@ -571,6 +571,10 @@ def fasterrcnn_resnet50_fpn(
 
 
 @register_model()
+@handle_legacy_interface(
+    weights=("pretrained", FasterRCNN_ResNet50_FPN_V2_Weights.COCO_V1),
+    weights_backbone=("pretrained_backbone", ResNet50_Weights.IMAGENET1K_V1),
+)
 def fasterrcnn_resnet50_fpn_v2(
     *,
     weights: Optional[FasterRCNN_ResNet50_FPN_V2_Weights] = None,

torchvision/models/detection/mask_rcnn.py

@@ -505,6 +505,10 @@ def maskrcnn_resnet50_fpn(
 
 
 @register_model()
+@handle_legacy_interface(
+    weights=("pretrained", MaskRCNN_ResNet50_FPN_V2_Weights.COCO_V1),
+    weights_backbone=("pretrained_backbone", ResNet50_Weights.IMAGENET1K_V1),
+)
 def maskrcnn_resnet50_fpn_v2(
     *,
     weights: Optional[MaskRCNN_ResNet50_FPN_V2_Weights] = None,

torchvision/models/detection/retinanet.py

@@ -819,6 +819,10 @@ def retinanet_resnet50_fpn(
 
 
 @register_model()
+@handle_legacy_interface(
+    weights=("pretrained", RetinaNet_ResNet50_FPN_V2_Weights.COCO_V1),
+    weights_backbone=("pretrained_backbone", ResNet50_Weights.IMAGENET1K_V1),
+)
 def retinanet_resnet50_fpn_v2(
     *,
     weights: Optional[RetinaNet_ResNet50_FPN_V2_Weights] = None,

torchvision/models/quantization/resnet.py

@@ -421,6 +421,14 @@ def resnext101_32x8d(
 
 
 @register_model(name="quantized_resnext101_64x4d")
+@handle_legacy_interface(
+    weights=(
+        "pretrained",
+        lambda kwargs: ResNeXt101_64X4D_QuantizedWeights.IMAGENET1K_FBGEMM_V1
+        if kwargs.get("quantize", False)
+        else ResNeXt101_64X4D_Weights.IMAGENET1K_V1,
+    )
+)
 def resnext101_64x4d(
     *,
     weights: Optional[Union[ResNeXt101_64X4D_QuantizedWeights, ResNeXt101_64X4D_Weights]] = None,

torchvision/models/quantization/shufflenetv2.py

@@ -312,6 +312,14 @@ def shufflenet_v2_x1_0(
 
 
 @register_model(name="quantized_shufflenet_v2_x1_5")
+@handle_legacy_interface(
+    weights=(
+        "pretrained",
+        lambda kwargs: ShuffleNet_V2_X1_5_QuantizedWeights.IMAGENET1K_FBGEMM_V1
+        if kwargs.get("quantize", False)
+        else ShuffleNet_V2_X1_5_Weights.IMAGENET1K_V1,
+    )
+)
 def shufflenet_v2_x1_5(
     *,
     weights: Optional[Union[ShuffleNet_V2_X1_5_QuantizedWeights, ShuffleNet_V2_X1_5_Weights]] = None,
@@ -358,6 +366,14 @@ def shufflenet_v2_x1_5(
 
 
 @register_model(name="quantized_shufflenet_v2_x2_0")
+@handle_legacy_interface(
+    weights=(
+        "pretrained",
+        lambda kwargs: ShuffleNet_V2_X2_0_QuantizedWeights.IMAGENET1K_FBGEMM_V1
+        if kwargs.get("quantize", False)
+        else ShuffleNet_V2_X2_0_Weights.IMAGENET1K_V1,
+    )
+)
 def shufflenet_v2_x2_0(
     *,
     weights: Optional[Union[ShuffleNet_V2_X2_0_QuantizedWeights, ShuffleNet_V2_X2_0_Weights]] = None,

torchvision/models/resnet.py

@@ -854,6 +854,7 @@ def resnext101_32x8d(
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", ResNeXt101_64X4D_Weights.IMAGENET1K_V1))
 def resnext101_64x4d(
     *, weights: Optional[ResNeXt101_64X4D_Weights] = None, progress: bool = True, **kwargs: Any
 ) -> ResNet:

torchvision/models/swin_transformer.py

@@ -12,7 +12,7 @@ from ..transforms._presets import ImageClassification, InterpolationMode
 from ..utils import _log_api_usage_once
 from ._api import register_model, Weights, WeightsEnum
 from ._meta import _IMAGENET_CATEGORIES
-from ._utils import _ovewrite_named_param
+from ._utils import _ovewrite_named_param, handle_legacy_interface
 
 
 __all__ = [
@@ -782,6 +782,7 @@ class Swin_V2_B_Weights(WeightsEnum):
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", Swin_T_Weights.IMAGENET1K_V1))
 def swin_t(*, weights: Optional[Swin_T_Weights] = None, progress: bool = True, **kwargs: Any) -> SwinTransformer:
     """
     Constructs a swin_tiny architecture from
@@ -819,6 +820,7 @@ def swin_t(*, weights: Optional[Swin_T_Weights] = None, progress: bool = True, *
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", Swin_S_Weights.IMAGENET1K_V1))
 def swin_s(*, weights: Optional[Swin_S_Weights] = None, progress: bool = True, **kwargs: Any) -> SwinTransformer:
     """
     Constructs a swin_small architecture from
@@ -856,6 +858,7 @@ def swin_s(*, weights: Optional[Swin_S_Weights] = None, progress: bool = True, *
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", Swin_B_Weights.IMAGENET1K_V1))
 def swin_b(*, weights: Optional[Swin_B_Weights] = None, progress: bool = True, **kwargs: Any) -> SwinTransformer:
     """
     Constructs a swin_base architecture from
@@ -893,6 +896,7 @@ def swin_b(*, weights: Optional[Swin_B_Weights] = None, progress: bool = True, *
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", Swin_V2_T_Weights.IMAGENET1K_V1))
 def swin_v2_t(*, weights: Optional[Swin_V2_T_Weights] = None, progress: bool = True, **kwargs: Any) -> SwinTransformer:
     """
     Constructs a swin_v2_tiny architecture from
@@ -932,6 +936,7 @@ def swin_v2_t(*, weights: Optional[Swin_V2_T_Weights] = None, progress: bool = T
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", Swin_V2_S_Weights.IMAGENET1K_V1))
 def swin_v2_s(*, weights: Optional[Swin_V2_S_Weights] = None, progress: bool = True, **kwargs: Any) -> SwinTransformer:
     """
     Constructs a swin_v2_small architecture from
@@ -971,6 +976,7 @@ def swin_v2_s(*, weights: Optional[Swin_V2_S_Weights] = None, progress: bool = T
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", Swin_V2_B_Weights.IMAGENET1K_V1))
 def swin_v2_b(*, weights: Optional[Swin_V2_B_Weights] = None, progress: bool = True, **kwargs: Any) -> SwinTransformer:
     """
     Constructs a swin_v2_base architecture from

torchvision/models/video/mvit.py

@@ -12,7 +12,7 @@ from ...transforms._presets import VideoClassification
 from ...utils import _log_api_usage_once
 from .._api import register_model, Weights, WeightsEnum
 from .._meta import _KINETICS400_CATEGORIES
-from .._utils import _ovewrite_named_param
+from .._utils import _ovewrite_named_param, handle_legacy_interface
 
 
 __all__ = [
@@ -661,6 +661,7 @@ class MViT_V2_S_Weights(WeightsEnum):
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", MViT_V1_B_Weights.KINETICS400_V1))
 def mvit_v1_b(*, weights: Optional[MViT_V1_B_Weights] = None, progress: bool = True, **kwargs: Any) -> MViT:
     """
     Constructs a base MViTV1 architecture from
@@ -756,6 +757,7 @@ def mvit_v1_b(*, weights: Optional[MViT_V1_B_Weights] = None, progress: bool = T
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", MViT_V2_S_Weights.KINETICS400_V1))
 def mvit_v2_s(*, weights: Optional[MViT_V2_S_Weights] = None, progress: bool = True, **kwargs: Any) -> MViT:
     """
     Constructs a small MViTV2 architecture from

torchvision/models/video/s3d.py

@@ -9,7 +9,7 @@ from ...transforms._presets import VideoClassification
 from ...utils import _log_api_usage_once
 from .._api import register_model, Weights, WeightsEnum
 from .._meta import _KINETICS400_CATEGORIES
-from .._utils import _ovewrite_named_param
+from .._utils import _ovewrite_named_param, handle_legacy_interface
 
 
 __all__ = [
@@ -181,6 +181,7 @@ class S3D_Weights(WeightsEnum):
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", S3D_Weights.KINETICS400_V1))
 def s3d(*, weights: Optional[S3D_Weights] = None, progress: bool = True, **kwargs: Any) -> S3D:
     """Construct Separable 3D CNN model.
 

torchvision/models/vision_transformer.py

@@ -733,6 +733,7 @@ def vit_l_32(*, weights: Optional[ViT_L_32_Weights] = None, progress: bool = Tru
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", None))
 def vit_h_14(*, weights: Optional[ViT_H_14_Weights] = None, progress: bool = True, **kwargs: Any) -> VisionTransformer:
     """
     Constructs a vit_h_14 architecture from

torchvision/prototype/models/depth/stereo/raft_stereo.py

@@ -6,6 +6,7 @@ import torch.nn.functional as F
 import torchvision.models.optical_flow.raft as raft
 from torch import Tensor
 from torchvision.models._api import register_model, WeightsEnum
+from torchvision.models._utils import handle_legacy_interface
 from torchvision.models.optical_flow._utils import grid_sample, make_coords_grid, upsample_flow
 from torchvision.models.optical_flow.raft import FlowHead, MotionEncoder, ResidualBlock
 from torchvision.ops import Conv2dNormActivation
@@ -618,6 +619,7 @@ class Raft_Stereo_Base_Weights(WeightsEnum):
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", None))
 def raft_stereo_realtime(
     *, weights: Optional[Raft_Stereo_Realtime_Weights] = None, progress=True, **kwargs
 ) -> RaftStereo:
@@ -678,6 +680,7 @@ def raft_stereo_realtime(
 
 
 @register_model()
+@handle_legacy_interface(weights=("pretrained", None))
 def raft_stereo_base(*, weights: Optional[Raft_Stereo_Base_Weights] = None, progress=True, **kwargs) -> RaftStereo:
     """RAFT-Stereo model from
     `RAFT-Stereo: Multilevel Recurrent Field Transforms for Stereo Matching <https://arxiv.org/abs/2109.07547>`_.