Expose check_inputs

Author: NicolasHug

torchvision/prototype/transforms/_geometry.py

@@ -35,7 +35,7 @@ def __init__(
 
         self.padding_mode = padding_mode
 
-    def _check_inputs(self, flat_inputs: List[Any]) -> None:
+    def check_inputs(self, flat_inputs: List[Any]) -> None:
         if not has_any(
             flat_inputs,
             PIL.Image.Image,

torchvision/transforms/v2/_geometry.py

@@ -366,7 +366,7 @@ def _call_kernel(self, functional: Callable, inpt: Any, *args: Any, **kwargs: An
     def _transform(self, inpt: Any, params: Dict[str, Any]) -> Any:
         return self._call_kernel(F.five_crop, inpt, self.size)
 
-    def _check_inputs(self, flat_inputs: List[Any]) -> None:
+    def check_inputs(self, flat_inputs: List[Any]) -> None:
         if has_any(flat_inputs, tv_tensors.BoundingBoxes, tv_tensors.Mask):
             raise TypeError(f"BoundingBoxes'es and Mask's are not supported by {type(self).__name__}()")
 
@@ -408,7 +408,7 @@ def _call_kernel(self, functional: Callable, inpt: Any, *args: Any, **kwargs: An
             )
         return super()._call_kernel(functional, inpt, *args, **kwargs)
 
-    def _check_inputs(self, flat_inputs: List[Any]) -> None:
+    def check_inputs(self, flat_inputs: List[Any]) -> None:
         if has_any(flat_inputs, tv_tensors.BoundingBoxes, tv_tensors.Mask):
             raise TypeError(f"BoundingBoxes'es and Mask's are not supported by {type(self).__name__}()")
 
@@ -1132,7 +1132,7 @@ def __init__(
         self.options = sampler_options
         self.trials = trials
 
-    def _check_inputs(self, flat_inputs: List[Any]) -> None:
+    def check_inputs(self, flat_inputs: List[Any]) -> None:
         if not (
             has_all(flat_inputs, tv_tensors.BoundingBoxes)
             and has_any(flat_inputs, PIL.Image.Image, tv_tensors.Image, is_pure_tensor)

torchvision/transforms/v2/_misc.py

@@ -99,7 +99,7 @@ def __init__(self, transformation_matrix: torch.Tensor, mean_vector: torch.Tenso
         self.transformation_matrix = transformation_matrix
         self.mean_vector = mean_vector
 
-    def _check_inputs(self, sample: Any) -> Any:
+    def check_inputs(self, sample: Any) -> Any:
         if has_any(sample, PIL.Image.Image):
             raise TypeError(f"{type(self).__name__}() does not support PIL images.")
 
@@ -157,7 +157,7 @@ def __init__(self, mean: Sequence[float], std: Sequence[float], inplace: bool =
         self.std = list(std)
         self.inplace = inplace
 
-    def _check_inputs(self, sample: Any) -> Any:
+    def check_inputs(self, sample: Any) -> Any:
         if has_any(sample, PIL.Image.Image):
             raise TypeError(f"{type(self).__name__}() does not support PIL images.")
 

torchvision/transforms/v2/_transform.py

@@ -24,17 +24,15 @@ def __init__(self) -> None:
         super().__init__()
         _log_api_usage_once(self)
 
-    def _check_inputs(self, flat_inputs: List[Any]) -> None:
+    def check_inputs(self, flat_inputs: List[Any]) -> None:
         pass
 
-    # This exists for BC. When v2 was introduced, this method was private. Now
-    # it's publicly exposed as `make_params()`. It cannot be exposed as
-    # `get_params()` because there is already a `get_params()` methods for v2
-    # transforms: it's the v1's `get_params()` that we have  to keep in order to
-    # guarantee 100% BC with v1. (It's defined in __init_subclass__ below).
-    def _get_params(self, flat_inputs: List[Any]) -> Dict[str, Any]:
-        return self.make_params(flat_inputs)
-
+    # When v2 was introduced, this method was private and called
+    # `_get_params()`. Now it's publicly exposed as `make_params()`. It cannot
+    # be exposed as `get_params()` because there is already a `get_params()`
+    # methods for v2 transforms: it's the v1's `get_params()` that we have  to
+    # keep in order to guarantee 100% BC with v1. (It's defined in
+    # __init_subclass__ below).
     def make_params(self, flat_inputs: List[Any]) -> Dict[str, Any]:
         return dict()
 
@@ -48,7 +46,7 @@ def _transform(self, inpt: Any, params: Dict[str, Any]) -> Any:
     def forward(self, *inputs: Any) -> Any:
         flat_inputs, spec = tree_flatten(inputs if len(inputs) > 1 else inputs[0])
 
-        self._check_inputs(flat_inputs)
+        self.check_inputs(flat_inputs)
 
         needs_transform_list = self._needs_transform_list(flat_inputs)
         params = self.make_params(
@@ -161,12 +159,12 @@ def __init__(self, p: float = 0.5) -> None:
     def forward(self, *inputs: Any) -> Any:
         # We need to almost duplicate `Transform.forward()` here since we always want to check the inputs, but return
         # early afterwards in case the random check triggers. The same result could be achieved by calling
-        # `super().forward()` after the random check, but that would call `self._check_inputs` twice.
+        # `super().forward()` after the random check, but that would call `self.check_inputs` twice.
 
         inputs = inputs if len(inputs) > 1 else inputs[0]
         flat_inputs, spec = tree_flatten(inputs)
 
-        self._check_inputs(flat_inputs)
+        self.check_inputs(flat_inputs)
 
         if torch.rand(1) >= self.p:
             return inputs