From 4e7ab9dc1df6b5df48717954ff2374358bf757a3 Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Mon, 29 Nov 2021 22:35:47 +0000
Subject: [PATCH 1/9] add api usage log to functional transforms

---
 torchvision/transforms/functional.py | 37 ++++++++++++++++++++++++++++
 1 file changed, 37 insertions(+)

diff --git a/torchvision/transforms/functional.py b/torchvision/transforms/functional.py
index bd5b170626e..7f427736033 100644
--- a/torchvision/transforms/functional.py
+++ b/torchvision/transforms/functional.py
@@ -14,6 +14,7 @@
 except ImportError:
     accimage = None
 
+from ..utils import _log_api_usage_once
 from . import functional_pil as F_pil
 from . import functional_tensor as F_t
 
@@ -56,6 +57,7 @@ def _interpolation_modes_from_int(i: int) -> InterpolationMode:
 }
 
 _is_pil_image = F_pil._is_pil_image
+_MODULE = "torchvision.transforms.functional"
 
 
 def get_image_size(img: Tensor) -> List[int]:
@@ -67,6 +69,7 @@ def get_image_size(img: Tensor) -> List[int]:
     Returns:
         List[int]: The image size.
     """
+    _log_api_usage_once(f"{_MODULE}.get_image_size")
     if isinstance(img, torch.Tensor):
         return F_t.get_image_size(img)
 
@@ -82,6 +85,7 @@ def get_image_num_channels(img: Tensor) -> int:
     Returns:
         int: The number of channels.
     """
+    _log_api_usage_once(f"{_MODULE}.get_image_num_channels")
     if isinstance(img, torch.Tensor):
         return F_t.get_image_num_channels(img)
 
@@ -110,6 +114,7 @@ def to_tensor(pic):
     Returns:
         Tensor: Converted image.
     """
+    _log_api_usage_once(f"{_MODULE}.to_tensor")
     if not (F_pil._is_pil_image(pic) or _is_numpy(pic)):
         raise TypeError(f"pic should be PIL Image or ndarray. Got {type(pic)}")
 
@@ -166,6 +171,7 @@ def pil_to_tensor(pic):
     Returns:
         Tensor: Converted image.
     """
+    _log_api_usage_once(f"{_MODULE}.pil_to_tensor")
     if not F_pil._is_pil_image(pic):
         raise TypeError(f"pic should be PIL Image. Got {type(pic)}")
 
@@ -205,6 +211,7 @@ def convert_image_dtype(image: torch.Tensor, dtype: torch.dtype = torch.float) -
             overflow errors since the floating point ``dtype`` cannot store consecutive integers over the whole range
             of the integer ``dtype``.
     """
+    _log_api_usage_once(f"{_MODULE}.convert_image_dtype")
     if not isinstance(image, torch.Tensor):
         raise TypeError("Input img should be Tensor Image")
 
@@ -225,6 +232,7 @@ def to_pil_image(pic, mode=None):
     Returns:
         PIL Image: Image converted to PIL Image.
     """
+    _log_api_usage_once(f"{_MODULE}.to_pil_image")
     if not (isinstance(pic, torch.Tensor) or isinstance(pic, np.ndarray)):
         raise TypeError(f"pic should be Tensor or ndarray. Got {type(pic)}.")
 
@@ -322,6 +330,7 @@ def normalize(tensor: Tensor, mean: List[float], std: List[float], inplace: bool
     Returns:
         Tensor: Normalized Tensor image.
     """
+    _log_api_usage_once(f"{_MODULE}.normalize")
     if not isinstance(tensor, torch.Tensor):
         raise TypeError(f"Input tensor should be a torch tensor. Got {type(tensor)}.")
 
@@ -401,6 +410,7 @@ def resize(
     Returns:
         PIL Image or Tensor: Resized image.
     """
+    _log_api_usage_once(f"{_MODULE}.resize")
     # Backward compatibility with integer value
     if isinstance(interpolation, int):
         warnings.warn(
@@ -422,6 +432,7 @@ def resize(
 
 
 def scale(*args, **kwargs):
+    _log_api_usage_once(f"{_MODULE}.scale")
     warnings.warn("The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.")
     return resize(*args, **kwargs)
 
@@ -467,6 +478,7 @@ def pad(img: Tensor, padding: List[int], fill: int = 0, padding_mode: str = "con
     Returns:
         PIL Image or Tensor: Padded image.
     """
+    _log_api_usage_once(f"{_MODULE}.pad")
     if not isinstance(img, torch.Tensor):
         return F_pil.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
 
@@ -490,6 +502,7 @@ def crop(img: Tensor, top: int, left: int, height: int, width: int) -> Tensor:
         PIL Image or Tensor: Cropped image.
     """
 
+    _log_api_usage_once(f"{_MODULE}.crop")
     if not isinstance(img, torch.Tensor):
         return F_pil.crop(img, top, left, height, width)
 
@@ -510,6 +523,7 @@ def center_crop(img: Tensor, output_size: List[int]) -> Tensor:
     Returns:
         PIL Image or Tensor: Cropped image.
     """
+    _log_api_usage_once(f"{_MODULE}.center_crop")
     if isinstance(output_size, numbers.Number):
         output_size = (int(output_size), int(output_size))
     elif isinstance(output_size, (tuple, list)) and len(output_size) == 1:
@@ -566,6 +580,7 @@ def resized_crop(
     Returns:
         PIL Image or Tensor: Cropped image.
     """
+    _log_api_usage_once(f"{_MODULE}.resized_crop")
     img = crop(img, top, left, height, width)
     img = resize(img, size, interpolation)
     return img
@@ -583,6 +598,7 @@ def hflip(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor:  Horizontally flipped image.
     """
+    _log_api_usage_once(f"{_MODULE}.hflip")
     if not isinstance(img, torch.Tensor):
         return F_pil.hflip(img)
 
@@ -648,6 +664,7 @@ def perspective(
     Returns:
         PIL Image or Tensor: transformed Image.
     """
+    _log_api_usage_once(f"{_MODULE}.perspective")
 
     coeffs = _get_perspective_coeffs(startpoints, endpoints)
 
@@ -681,6 +698,7 @@ def vflip(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor:  Vertically flipped image.
     """
+    _log_api_usage_once(f"{_MODULE}.vflip")
     if not isinstance(img, torch.Tensor):
         return F_pil.vflip(img)
 
@@ -706,6 +724,7 @@ def five_crop(img: Tensor, size: List[int]) -> Tuple[Tensor, Tensor, Tensor, Ten
        tuple: tuple (tl, tr, bl, br, center)
        Corresponding top left, top right, bottom left, bottom right and center crop.
     """
+    _log_api_usage_once(f"{_MODULE}.five_crop")
     if isinstance(size, numbers.Number):
         size = (int(size), int(size))
     elif isinstance(size, (tuple, list)) and len(size) == 1:
@@ -753,6 +772,7 @@ def ten_crop(img: Tensor, size: List[int], vertical_flip: bool = False) -> List[
         Corresponding top left, top right, bottom left, bottom right and
         center crop and same for the flipped image.
     """
+    _log_api_usage_once(f"{_MODULE}.ten_crop")
     if isinstance(size, numbers.Number):
         size = (int(size), int(size))
     elif isinstance(size, (tuple, list)) and len(size) == 1:
@@ -786,6 +806,7 @@ def adjust_brightness(img: Tensor, brightness_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Brightness adjusted image.
     """
+    _log_api_usage_once(f"{_MODULE}.adjust_brightness")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_brightness(img, brightness_factor)
 
@@ -806,6 +827,7 @@ def adjust_contrast(img: Tensor, contrast_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Contrast adjusted image.
     """
+    _log_api_usage_once(f"{_MODULE}.adjust_contrast")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_contrast(img, contrast_factor)
 
@@ -826,6 +848,7 @@ def adjust_saturation(img: Tensor, saturation_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Saturation adjusted image.
     """
+    _log_api_usage_once(f"{_MODULE}.adjust_saturation")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_saturation(img, saturation_factor)
 
@@ -860,6 +883,7 @@ def adjust_hue(img: Tensor, hue_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Hue adjusted image.
     """
+    _log_api_usage_once(f"{_MODULE}.adjust_hue")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_hue(img, hue_factor)
 
@@ -891,6 +915,7 @@ def adjust_gamma(img: Tensor, gamma: float, gain: float = 1) -> Tensor:
     Returns:
         PIL Image or Tensor: Gamma correction adjusted image.
     """
+    _log_api_usage_once(f"{_MODULE}.adjust_gamma")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_gamma(img, gamma, gain)
 
@@ -987,6 +1012,7 @@ def rotate(
     .. _filters: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#filters
 
     """
+    _log_api_usage_once(f"{_MODULE}.rotate")
     if resample is not None:
         warnings.warn(
             "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1067,6 +1093,7 @@ def affine(
     Returns:
         PIL Image or Tensor: Transformed image.
     """
+    _log_api_usage_once(f"{_MODULE}.affine")
     if resample is not None:
         warnings.warn(
             "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1151,6 +1178,7 @@ def to_grayscale(img, num_output_channels=1):
         - if num_output_channels = 1 : returned image is single channel
         - if num_output_channels = 3 : returned image is 3 channel with r = g = b
     """
+    _log_api_usage_once(f"{_MODULE}.to_grayscale")
     if isinstance(img, Image.Image):
         return F_pil.to_grayscale(img, num_output_channels)
 
@@ -1176,6 +1204,7 @@ def rgb_to_grayscale(img: Tensor, num_output_channels: int = 1) -> Tensor:
         - if num_output_channels = 1 : returned image is single channel
         - if num_output_channels = 3 : returned image is 3 channel with r = g = b
     """
+    _log_api_usage_once(f"{_MODULE}.rgb_to_grayscale")
     if not isinstance(img, torch.Tensor):
         return F_pil.to_grayscale(img, num_output_channels)
 
@@ -1198,6 +1227,7 @@ def erase(img: Tensor, i: int, j: int, h: int, w: int, v: Tensor, inplace: bool
     Returns:
         Tensor Image: Erased image.
     """
+    _log_api_usage_once(f"{_MODULE}.erase")
     if not isinstance(img, torch.Tensor):
         raise TypeError(f"img should be Tensor Image. Got {type(img)}")
 
@@ -1234,6 +1264,7 @@ def gaussian_blur(img: Tensor, kernel_size: List[int], sigma: Optional[List[floa
     Returns:
         PIL Image or Tensor: Gaussian Blurred version of the image.
     """
+    _log_api_usage_once(f"{_MODULE}.gaussian_blur")
     if not isinstance(kernel_size, (int, list, tuple)):
         raise TypeError(f"kernel_size should be int or a sequence of integers. Got {type(kernel_size)}")
     if isinstance(kernel_size, int):
@@ -1285,6 +1316,7 @@ def invert(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: Color inverted image.
     """
+    _log_api_usage_once(f"{_MODULE}.invert")
     if not isinstance(img, torch.Tensor):
         return F_pil.invert(img)
 
@@ -1304,6 +1336,7 @@ def posterize(img: Tensor, bits: int) -> Tensor:
     Returns:
         PIL Image or Tensor: Posterized image.
     """
+    _log_api_usage_once(f"{_MODULE}.posterize")
     if not (0 <= bits <= 8):
         raise ValueError(f"The number if bits should be between 0 and 8. Got {bits}")
 
@@ -1325,6 +1358,7 @@ def solarize(img: Tensor, threshold: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Solarized image.
     """
+    _log_api_usage_once(f"{_MODULE}.solarize")
     if not isinstance(img, torch.Tensor):
         return F_pil.solarize(img, threshold)
 
@@ -1345,6 +1379,7 @@ def adjust_sharpness(img: Tensor, sharpness_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Sharpness adjusted image.
     """
+    _log_api_usage_once(f"{_MODULE}.adjust_sharpness")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_sharpness(img, sharpness_factor)
 
@@ -1365,6 +1400,7 @@ def autocontrast(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: An image that was autocontrasted.
     """
+    _log_api_usage_once(f"{_MODULE}.autocontrast")
     if not isinstance(img, torch.Tensor):
         return F_pil.autocontrast(img)
 
@@ -1386,6 +1422,7 @@ def equalize(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: An image that was equalized.
     """
+    _log_api_usage_once(f"{_MODULE}.equalize")
     if not isinstance(img, torch.Tensor):
         return F_pil.equalize(img)
 

From 4649e4d20e6aeb792b296622716cb49583cc6a65 Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Mon, 29 Nov 2021 22:51:40 +0000
Subject: [PATCH 2/9] add log to transforms

---
 torchvision/transforms/transforms.py | 24 ++++++++++++++++++++++++
 1 file changed, 24 insertions(+)

diff --git a/torchvision/transforms/transforms.py b/torchvision/transforms/transforms.py
index a409ff3cbb8..dab67e1abea 100644
--- a/torchvision/transforms/transforms.py
+++ b/torchvision/transforms/transforms.py
@@ -13,6 +13,7 @@
 except ImportError:
     accimage = None
 
+from ..utils import _log_api_usage_once
 from . import functional as F
 from .functional import InterpolationMode, _interpolation_modes_from_int
 
@@ -87,6 +88,7 @@ class Compose:
     """
 
     def __init__(self, transforms):
+        _log_api_usage_once(self)
         self.transforms = transforms
 
     def __call__(self, img):
@@ -462,6 +464,7 @@ class Lambda:
     """
 
     def __init__(self, lambd):
+        _log_api_usage_once(self)
         if not callable(lambd):
             raise TypeError(f"Argument lambd should be callable, got {repr(type(lambd).__name__)}")
         self.lambd = lambd
@@ -481,6 +484,7 @@ class RandomTransforms:
     """
 
     def __init__(self, transforms):
+        _log_api_usage_once(self)
         if not isinstance(transforms, Sequence):
             raise TypeError("Argument transforms should be a sequence")
         self.transforms = transforms
@@ -519,6 +523,7 @@ class RandomApply(torch.nn.Module):
 
     def __init__(self, transforms, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.transforms = transforms
         self.p = p
 
@@ -639,6 +644,7 @@ def get_params(img: Tensor, output_size: Tuple[int, int]) -> Tuple[int, int, int
 
     def __init__(self, size, padding=None, pad_if_needed=False, fill=0, padding_mode="constant"):
         super().__init__()
+        _log_api_usage_once(self)
 
         self.size = tuple(_setup_size(size, error_msg="Please provide only two dimensions (h, w) for size."))
 
@@ -688,6 +694,7 @@ class RandomHorizontalFlip(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -718,6 +725,7 @@ class RandomVerticalFlip(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -755,6 +763,7 @@ class RandomPerspective(torch.nn.Module):
 
     def __init__(self, distortion_scale=0.5, p=0.5, interpolation=InterpolationMode.BILINEAR, fill=0):
         super().__init__()
+        _log_api_usage_once(self)
         self.p = p
 
         # Backward compatibility with integer value
@@ -867,6 +876,7 @@ class RandomResizedCrop(torch.nn.Module):
 
     def __init__(self, size, scale=(0.08, 1.0), ratio=(3.0 / 4.0, 4.0 / 3.0), interpolation=InterpolationMode.BILINEAR):
         super().__init__()
+        _log_api_usage_once(self)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
 
         if not isinstance(scale, Sequence):
@@ -1081,6 +1091,7 @@ class LinearTransformation(torch.nn.Module):
 
     def __init__(self, transformation_matrix, mean_vector):
         super().__init__()
+        _log_api_usage_once(self)
         if transformation_matrix.size(0) != transformation_matrix.size(1):
             raise ValueError(
                 "transformation_matrix should be square. Got "
@@ -1159,6 +1170,7 @@ class ColorJitter(torch.nn.Module):
 
     def __init__(self, brightness=0, contrast=0, saturation=0, hue=0):
         super().__init__()
+        _log_api_usage_once(self)
         self.brightness = self._check_input(brightness, "brightness")
         self.contrast = self._check_input(contrast, "contrast")
         self.saturation = self._check_input(saturation, "saturation")
@@ -1281,6 +1293,7 @@ def __init__(
         self, degrees, interpolation=InterpolationMode.NEAREST, expand=False, center=None, fill=0, resample=None
     ):
         super().__init__()
+        _log_api_usage_once(self)
         if resample is not None:
             warnings.warn(
                 "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1401,6 +1414,7 @@ def __init__(
         resample=None,
     ):
         super().__init__()
+        _log_api_usage_once(self)
         if resample is not None:
             warnings.warn(
                 "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1545,6 +1559,7 @@ class Grayscale(torch.nn.Module):
 
     def __init__(self, num_output_channels=1):
         super().__init__()
+        _log_api_usage_once(self)
         self.num_output_channels = num_output_channels
 
     def forward(self, img):
@@ -1579,6 +1594,7 @@ class RandomGrayscale(torch.nn.Module):
 
     def __init__(self, p=0.1):
         super().__init__()
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -1628,6 +1644,7 @@ class RandomErasing(torch.nn.Module):
 
     def __init__(self, p=0.5, scale=(0.02, 0.33), ratio=(0.3, 3.3), value=0, inplace=False):
         super().__init__()
+        _log_api_usage_once(self)
         if not isinstance(value, (numbers.Number, str, tuple, list)):
             raise TypeError("Argument value should be either a number or str or a sequence")
         if isinstance(value, str) and value != "random":
@@ -1751,6 +1768,7 @@ class GaussianBlur(torch.nn.Module):
 
     def __init__(self, kernel_size, sigma=(0.1, 2.0)):
         super().__init__()
+        _log_api_usage_once(self)
         self.kernel_size = _setup_size(kernel_size, "Kernel size should be a tuple/list of two integers")
         for ks in self.kernel_size:
             if ks <= 0 or ks % 2 == 0:
@@ -1842,6 +1860,7 @@ class RandomInvert(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -1873,6 +1892,7 @@ class RandomPosterize(torch.nn.Module):
 
     def __init__(self, bits, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.bits = bits
         self.p = p
 
@@ -1905,6 +1925,7 @@ class RandomSolarize(torch.nn.Module):
 
     def __init__(self, threshold, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.threshold = threshold
         self.p = p
 
@@ -1937,6 +1958,7 @@ class RandomAdjustSharpness(torch.nn.Module):
 
     def __init__(self, sharpness_factor, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.sharpness_factor = sharpness_factor
         self.p = p
 
@@ -1968,6 +1990,7 @@ class RandomAutocontrast(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -1998,6 +2021,7 @@ class RandomEqualize(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):

From 470dadf3a63f514567b3787bfcff8ad102d29cb0 Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Mon, 29 Nov 2021 23:21:39 +0000
Subject: [PATCH 3/9] fix for scriptablity

---
 torchvision/transforms/functional.py | 71 ++++++++++++++--------------
 1 file changed, 35 insertions(+), 36 deletions(-)

diff --git a/torchvision/transforms/functional.py b/torchvision/transforms/functional.py
index 7f427736033..f24fe070726 100644
--- a/torchvision/transforms/functional.py
+++ b/torchvision/transforms/functional.py
@@ -57,7 +57,6 @@ def _interpolation_modes_from_int(i: int) -> InterpolationMode:
 }
 
 _is_pil_image = F_pil._is_pil_image
-_MODULE = "torchvision.transforms.functional"
 
 
 def get_image_size(img: Tensor) -> List[int]:
@@ -69,7 +68,7 @@ def get_image_size(img: Tensor) -> List[int]:
     Returns:
         List[int]: The image size.
     """
-    _log_api_usage_once(f"{_MODULE}.get_image_size")
+    _log_api_usage_once("torchvision.transforms.functional.get_image_size")
     if isinstance(img, torch.Tensor):
         return F_t.get_image_size(img)
 
@@ -85,7 +84,7 @@ def get_image_num_channels(img: Tensor) -> int:
     Returns:
         int: The number of channels.
     """
-    _log_api_usage_once(f"{_MODULE}.get_image_num_channels")
+    _log_api_usage_once("torchvision.transforms.functional.get_image_num_channels")
     if isinstance(img, torch.Tensor):
         return F_t.get_image_num_channels(img)
 
@@ -114,7 +113,7 @@ def to_tensor(pic):
     Returns:
         Tensor: Converted image.
     """
-    _log_api_usage_once(f"{_MODULE}.to_tensor")
+    _log_api_usage_once("torchvision.transforms.functional.to_tensor")
     if not (F_pil._is_pil_image(pic) or _is_numpy(pic)):
         raise TypeError(f"pic should be PIL Image or ndarray. Got {type(pic)}")
 
@@ -171,7 +170,7 @@ def pil_to_tensor(pic):
     Returns:
         Tensor: Converted image.
     """
-    _log_api_usage_once(f"{_MODULE}.pil_to_tensor")
+    _log_api_usage_once("torchvision.transforms.functional.pil_to_tensor")
     if not F_pil._is_pil_image(pic):
         raise TypeError(f"pic should be PIL Image. Got {type(pic)}")
 
@@ -211,7 +210,7 @@ def convert_image_dtype(image: torch.Tensor, dtype: torch.dtype = torch.float) -
             overflow errors since the floating point ``dtype`` cannot store consecutive integers over the whole range
             of the integer ``dtype``.
     """
-    _log_api_usage_once(f"{_MODULE}.convert_image_dtype")
+    _log_api_usage_once("torchvision.transforms.functional.convert_image_dtype")
     if not isinstance(image, torch.Tensor):
         raise TypeError("Input img should be Tensor Image")
 
@@ -232,7 +231,7 @@ def to_pil_image(pic, mode=None):
     Returns:
         PIL Image: Image converted to PIL Image.
     """
-    _log_api_usage_once(f"{_MODULE}.to_pil_image")
+    _log_api_usage_once("torchvision.transforms.functional.to_pil_image")
     if not (isinstance(pic, torch.Tensor) or isinstance(pic, np.ndarray)):
         raise TypeError(f"pic should be Tensor or ndarray. Got {type(pic)}.")
 
@@ -330,7 +329,7 @@ def normalize(tensor: Tensor, mean: List[float], std: List[float], inplace: bool
     Returns:
         Tensor: Normalized Tensor image.
     """
-    _log_api_usage_once(f"{_MODULE}.normalize")
+    _log_api_usage_once("torchvision.transforms.functional.normalize")
     if not isinstance(tensor, torch.Tensor):
         raise TypeError(f"Input tensor should be a torch tensor. Got {type(tensor)}.")
 
@@ -410,7 +409,7 @@ def resize(
     Returns:
         PIL Image or Tensor: Resized image.
     """
-    _log_api_usage_once(f"{_MODULE}.resize")
+    _log_api_usage_once("torchvision.transforms.functional.resize")
     # Backward compatibility with integer value
     if isinstance(interpolation, int):
         warnings.warn(
@@ -432,7 +431,7 @@ def resize(
 
 
 def scale(*args, **kwargs):
-    _log_api_usage_once(f"{_MODULE}.scale")
+    _log_api_usage_once("torchvision.transforms.functional.scale")
     warnings.warn("The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.")
     return resize(*args, **kwargs)
 
@@ -478,7 +477,7 @@ def pad(img: Tensor, padding: List[int], fill: int = 0, padding_mode: str = "con
     Returns:
         PIL Image or Tensor: Padded image.
     """
-    _log_api_usage_once(f"{_MODULE}.pad")
+    _log_api_usage_once("torchvision.transforms.functional.pad")
     if not isinstance(img, torch.Tensor):
         return F_pil.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
 
@@ -502,7 +501,7 @@ def crop(img: Tensor, top: int, left: int, height: int, width: int) -> Tensor:
         PIL Image or Tensor: Cropped image.
     """
 
-    _log_api_usage_once(f"{_MODULE}.crop")
+    _log_api_usage_once("torchvision.transforms.functional.crop")
     if not isinstance(img, torch.Tensor):
         return F_pil.crop(img, top, left, height, width)
 
@@ -523,7 +522,7 @@ def center_crop(img: Tensor, output_size: List[int]) -> Tensor:
     Returns:
         PIL Image or Tensor: Cropped image.
     """
-    _log_api_usage_once(f"{_MODULE}.center_crop")
+    _log_api_usage_once("torchvision.transforms.functional.center_crop")
     if isinstance(output_size, numbers.Number):
         output_size = (int(output_size), int(output_size))
     elif isinstance(output_size, (tuple, list)) and len(output_size) == 1:
@@ -580,7 +579,7 @@ def resized_crop(
     Returns:
         PIL Image or Tensor: Cropped image.
     """
-    _log_api_usage_once(f"{_MODULE}.resized_crop")
+    _log_api_usage_once("torchvision.transforms.functional.resized_crop")
     img = crop(img, top, left, height, width)
     img = resize(img, size, interpolation)
     return img
@@ -598,7 +597,7 @@ def hflip(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor:  Horizontally flipped image.
     """
-    _log_api_usage_once(f"{_MODULE}.hflip")
+    _log_api_usage_once("torchvision.transforms.functional.hflip")
     if not isinstance(img, torch.Tensor):
         return F_pil.hflip(img)
 
@@ -664,7 +663,7 @@ def perspective(
     Returns:
         PIL Image or Tensor: transformed Image.
     """
-    _log_api_usage_once(f"{_MODULE}.perspective")
+    _log_api_usage_once("torchvision.transforms.functional.perspective")
 
     coeffs = _get_perspective_coeffs(startpoints, endpoints)
 
@@ -698,7 +697,7 @@ def vflip(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor:  Vertically flipped image.
     """
-    _log_api_usage_once(f"{_MODULE}.vflip")
+    _log_api_usage_once("torchvision.transforms.functional.vflip")
     if not isinstance(img, torch.Tensor):
         return F_pil.vflip(img)
 
@@ -724,7 +723,7 @@ def five_crop(img: Tensor, size: List[int]) -> Tuple[Tensor, Tensor, Tensor, Ten
        tuple: tuple (tl, tr, bl, br, center)
        Corresponding top left, top right, bottom left, bottom right and center crop.
     """
-    _log_api_usage_once(f"{_MODULE}.five_crop")
+    _log_api_usage_once("torchvision.transforms.functional.five_crop")
     if isinstance(size, numbers.Number):
         size = (int(size), int(size))
     elif isinstance(size, (tuple, list)) and len(size) == 1:
@@ -772,7 +771,7 @@ def ten_crop(img: Tensor, size: List[int], vertical_flip: bool = False) -> List[
         Corresponding top left, top right, bottom left, bottom right and
         center crop and same for the flipped image.
     """
-    _log_api_usage_once(f"{_MODULE}.ten_crop")
+    _log_api_usage_once("torchvision.transforms.functional.ten_crop")
     if isinstance(size, numbers.Number):
         size = (int(size), int(size))
     elif isinstance(size, (tuple, list)) and len(size) == 1:
@@ -806,7 +805,7 @@ def adjust_brightness(img: Tensor, brightness_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Brightness adjusted image.
     """
-    _log_api_usage_once(f"{_MODULE}.adjust_brightness")
+    _log_api_usage_once("torchvision.transforms.functional.adjust_brightness")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_brightness(img, brightness_factor)
 
@@ -827,7 +826,7 @@ def adjust_contrast(img: Tensor, contrast_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Contrast adjusted image.
     """
-    _log_api_usage_once(f"{_MODULE}.adjust_contrast")
+    _log_api_usage_once("torchvision.transforms.functional.adjust_contrast")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_contrast(img, contrast_factor)
 
@@ -848,7 +847,7 @@ def adjust_saturation(img: Tensor, saturation_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Saturation adjusted image.
     """
-    _log_api_usage_once(f"{_MODULE}.adjust_saturation")
+    _log_api_usage_once("torchvision.transforms.functional.adjust_saturation")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_saturation(img, saturation_factor)
 
@@ -883,7 +882,7 @@ def adjust_hue(img: Tensor, hue_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Hue adjusted image.
     """
-    _log_api_usage_once(f"{_MODULE}.adjust_hue")
+    _log_api_usage_once("torchvision.transforms.functional.adjust_hue")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_hue(img, hue_factor)
 
@@ -915,7 +914,7 @@ def adjust_gamma(img: Tensor, gamma: float, gain: float = 1) -> Tensor:
     Returns:
         PIL Image or Tensor: Gamma correction adjusted image.
     """
-    _log_api_usage_once(f"{_MODULE}.adjust_gamma")
+    _log_api_usage_once("torchvision.transforms.functional.adjust_gamma")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_gamma(img, gamma, gain)
 
@@ -1012,7 +1011,7 @@ def rotate(
     .. _filters: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#filters
 
     """
-    _log_api_usage_once(f"{_MODULE}.rotate")
+    _log_api_usage_once("torchvision.transforms.functional.rotate")
     if resample is not None:
         warnings.warn(
             "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1093,7 +1092,7 @@ def affine(
     Returns:
         PIL Image or Tensor: Transformed image.
     """
-    _log_api_usage_once(f"{_MODULE}.affine")
+    _log_api_usage_once("torchvision.transforms.functional.affine")
     if resample is not None:
         warnings.warn(
             "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1178,7 +1177,7 @@ def to_grayscale(img, num_output_channels=1):
         - if num_output_channels = 1 : returned image is single channel
         - if num_output_channels = 3 : returned image is 3 channel with r = g = b
     """
-    _log_api_usage_once(f"{_MODULE}.to_grayscale")
+    _log_api_usage_once("torchvision.transforms.functional.to_grayscale")
     if isinstance(img, Image.Image):
         return F_pil.to_grayscale(img, num_output_channels)
 
@@ -1204,7 +1203,7 @@ def rgb_to_grayscale(img: Tensor, num_output_channels: int = 1) -> Tensor:
         - if num_output_channels = 1 : returned image is single channel
         - if num_output_channels = 3 : returned image is 3 channel with r = g = b
     """
-    _log_api_usage_once(f"{_MODULE}.rgb_to_grayscale")
+    _log_api_usage_once("torchvision.transforms.functional.rgb_to_grayscale")
     if not isinstance(img, torch.Tensor):
         return F_pil.to_grayscale(img, num_output_channels)
 
@@ -1227,7 +1226,7 @@ def erase(img: Tensor, i: int, j: int, h: int, w: int, v: Tensor, inplace: bool
     Returns:
         Tensor Image: Erased image.
     """
-    _log_api_usage_once(f"{_MODULE}.erase")
+    _log_api_usage_once("torchvision.transforms.functional.erase")
     if not isinstance(img, torch.Tensor):
         raise TypeError(f"img should be Tensor Image. Got {type(img)}")
 
@@ -1264,7 +1263,7 @@ def gaussian_blur(img: Tensor, kernel_size: List[int], sigma: Optional[List[floa
     Returns:
         PIL Image or Tensor: Gaussian Blurred version of the image.
     """
-    _log_api_usage_once(f"{_MODULE}.gaussian_blur")
+    _log_api_usage_once("torchvision.transforms.functional.gaussian_blur")
     if not isinstance(kernel_size, (int, list, tuple)):
         raise TypeError(f"kernel_size should be int or a sequence of integers. Got {type(kernel_size)}")
     if isinstance(kernel_size, int):
@@ -1316,7 +1315,7 @@ def invert(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: Color inverted image.
     """
-    _log_api_usage_once(f"{_MODULE}.invert")
+    _log_api_usage_once("torchvision.transforms.functional.invert")
     if not isinstance(img, torch.Tensor):
         return F_pil.invert(img)
 
@@ -1336,7 +1335,7 @@ def posterize(img: Tensor, bits: int) -> Tensor:
     Returns:
         PIL Image or Tensor: Posterized image.
     """
-    _log_api_usage_once(f"{_MODULE}.posterize")
+    _log_api_usage_once("torchvision.transforms.functional.posterize")
     if not (0 <= bits <= 8):
         raise ValueError(f"The number if bits should be between 0 and 8. Got {bits}")
 
@@ -1358,7 +1357,7 @@ def solarize(img: Tensor, threshold: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Solarized image.
     """
-    _log_api_usage_once(f"{_MODULE}.solarize")
+    _log_api_usage_once("torchvision.transforms.functional.solarize")
     if not isinstance(img, torch.Tensor):
         return F_pil.solarize(img, threshold)
 
@@ -1379,7 +1378,7 @@ def adjust_sharpness(img: Tensor, sharpness_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Sharpness adjusted image.
     """
-    _log_api_usage_once(f"{_MODULE}.adjust_sharpness")
+    _log_api_usage_once("torchvision.transforms.functional.adjust_sharpness")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_sharpness(img, sharpness_factor)
 
@@ -1400,7 +1399,7 @@ def autocontrast(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: An image that was autocontrasted.
     """
-    _log_api_usage_once(f"{_MODULE}.autocontrast")
+    _log_api_usage_once("torchvision.transforms.functional.autocontrast")
     if not isinstance(img, torch.Tensor):
         return F_pil.autocontrast(img)
 
@@ -1422,7 +1421,7 @@ def equalize(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: An image that was equalized.
     """
-    _log_api_usage_once(f"{_MODULE}.equalize")
+    _log_api_usage_once("torchvision.transforms.functional.equalize")
     if not isinstance(img, torch.Tensor):
         return F_pil.equalize(img)
 

From 37884a06a437f3254c06adbad08d170bcfbc52b7 Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Tue, 30 Nov 2021 00:07:22 +0000
Subject: [PATCH 4/9] skip Compose for scriptability

---
 torchvision/transforms/transforms.py | 1 -
 1 file changed, 1 deletion(-)

diff --git a/torchvision/transforms/transforms.py b/torchvision/transforms/transforms.py
index dab67e1abea..37c9a709b25 100644
--- a/torchvision/transforms/transforms.py
+++ b/torchvision/transforms/transforms.py
@@ -88,7 +88,6 @@ class Compose:
     """
 
     def __init__(self, transforms):
-        _log_api_usage_once(self)
         self.transforms = transforms
 
     def __call__(self, img):

From e3fd8bb71b927adb9dbc1feb54d8547102e26c96 Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Thu, 9 Dec 2021 23:51:08 +0000
Subject: [PATCH 5/9] follow the new policy

---
 torchvision/transforms/functional.py | 70 ++++++++++++++--------------
 torchvision/transforms/transforms.py | 57 +++++++++++++---------
 2 files changed, 70 insertions(+), 57 deletions(-)

diff --git a/torchvision/transforms/functional.py b/torchvision/transforms/functional.py
index f24fe070726..99f4ce7e203 100644
--- a/torchvision/transforms/functional.py
+++ b/torchvision/transforms/functional.py
@@ -68,7 +68,7 @@ def get_image_size(img: Tensor) -> List[int]:
     Returns:
         List[int]: The image size.
     """
-    _log_api_usage_once("torchvision.transforms.functional.get_image_size")
+    _log_api_usage_once("transforms", "get_image_size")
     if isinstance(img, torch.Tensor):
         return F_t.get_image_size(img)
 
@@ -84,7 +84,7 @@ def get_image_num_channels(img: Tensor) -> int:
     Returns:
         int: The number of channels.
     """
-    _log_api_usage_once("torchvision.transforms.functional.get_image_num_channels")
+    _log_api_usage_once("transforms", "get_image_num_channels")
     if isinstance(img, torch.Tensor):
         return F_t.get_image_num_channels(img)
 
@@ -113,7 +113,7 @@ def to_tensor(pic):
     Returns:
         Tensor: Converted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.to_tensor")
+    _log_api_usage_once("transforms", "to_tensor")
     if not (F_pil._is_pil_image(pic) or _is_numpy(pic)):
         raise TypeError(f"pic should be PIL Image or ndarray. Got {type(pic)}")
 
@@ -170,7 +170,7 @@ def pil_to_tensor(pic):
     Returns:
         Tensor: Converted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.pil_to_tensor")
+    _log_api_usage_once("transforms", "pil_to_tensor")
     if not F_pil._is_pil_image(pic):
         raise TypeError(f"pic should be PIL Image. Got {type(pic)}")
 
@@ -210,7 +210,7 @@ def convert_image_dtype(image: torch.Tensor, dtype: torch.dtype = torch.float) -
             overflow errors since the floating point ``dtype`` cannot store consecutive integers over the whole range
             of the integer ``dtype``.
     """
-    _log_api_usage_once("torchvision.transforms.functional.convert_image_dtype")
+    _log_api_usage_once("transforms", "convert_image_dtype")
     if not isinstance(image, torch.Tensor):
         raise TypeError("Input img should be Tensor Image")
 
@@ -231,7 +231,7 @@ def to_pil_image(pic, mode=None):
     Returns:
         PIL Image: Image converted to PIL Image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.to_pil_image")
+    _log_api_usage_once("transforms", "to_pil_image")
     if not (isinstance(pic, torch.Tensor) or isinstance(pic, np.ndarray)):
         raise TypeError(f"pic should be Tensor or ndarray. Got {type(pic)}.")
 
@@ -329,7 +329,7 @@ def normalize(tensor: Tensor, mean: List[float], std: List[float], inplace: bool
     Returns:
         Tensor: Normalized Tensor image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.normalize")
+    _log_api_usage_once("transforms", "normalize")
     if not isinstance(tensor, torch.Tensor):
         raise TypeError(f"Input tensor should be a torch tensor. Got {type(tensor)}.")
 
@@ -409,7 +409,7 @@ def resize(
     Returns:
         PIL Image or Tensor: Resized image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.resize")
+    _log_api_usage_once("transforms", "resize")
     # Backward compatibility with integer value
     if isinstance(interpolation, int):
         warnings.warn(
@@ -431,7 +431,7 @@ def resize(
 
 
 def scale(*args, **kwargs):
-    _log_api_usage_once("torchvision.transforms.functional.scale")
+    _log_api_usage_once("transforms", "scale")
     warnings.warn("The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.")
     return resize(*args, **kwargs)
 
@@ -477,7 +477,7 @@ def pad(img: Tensor, padding: List[int], fill: int = 0, padding_mode: str = "con
     Returns:
         PIL Image or Tensor: Padded image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.pad")
+    _log_api_usage_once("transforms", "pad")
     if not isinstance(img, torch.Tensor):
         return F_pil.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
 
@@ -501,7 +501,7 @@ def crop(img: Tensor, top: int, left: int, height: int, width: int) -> Tensor:
         PIL Image or Tensor: Cropped image.
     """
 
-    _log_api_usage_once("torchvision.transforms.functional.crop")
+    _log_api_usage_once("transforms", "crop")
     if not isinstance(img, torch.Tensor):
         return F_pil.crop(img, top, left, height, width)
 
@@ -522,7 +522,7 @@ def center_crop(img: Tensor, output_size: List[int]) -> Tensor:
     Returns:
         PIL Image or Tensor: Cropped image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.center_crop")
+    _log_api_usage_once("transforms", "center_crop")
     if isinstance(output_size, numbers.Number):
         output_size = (int(output_size), int(output_size))
     elif isinstance(output_size, (tuple, list)) and len(output_size) == 1:
@@ -579,7 +579,7 @@ def resized_crop(
     Returns:
         PIL Image or Tensor: Cropped image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.resized_crop")
+    _log_api_usage_once("transforms", "resized_crop")
     img = crop(img, top, left, height, width)
     img = resize(img, size, interpolation)
     return img
@@ -597,7 +597,7 @@ def hflip(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor:  Horizontally flipped image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.hflip")
+    _log_api_usage_once("transforms", "hflip")
     if not isinstance(img, torch.Tensor):
         return F_pil.hflip(img)
 
@@ -663,7 +663,7 @@ def perspective(
     Returns:
         PIL Image or Tensor: transformed Image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.perspective")
+    _log_api_usage_once("transforms", "perspective")
 
     coeffs = _get_perspective_coeffs(startpoints, endpoints)
 
@@ -697,7 +697,7 @@ def vflip(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor:  Vertically flipped image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.vflip")
+    _log_api_usage_once("transforms", "vflip")
     if not isinstance(img, torch.Tensor):
         return F_pil.vflip(img)
 
@@ -723,7 +723,7 @@ def five_crop(img: Tensor, size: List[int]) -> Tuple[Tensor, Tensor, Tensor, Ten
        tuple: tuple (tl, tr, bl, br, center)
        Corresponding top left, top right, bottom left, bottom right and center crop.
     """
-    _log_api_usage_once("torchvision.transforms.functional.five_crop")
+    _log_api_usage_once("transforms", "five_crop")
     if isinstance(size, numbers.Number):
         size = (int(size), int(size))
     elif isinstance(size, (tuple, list)) and len(size) == 1:
@@ -771,7 +771,7 @@ def ten_crop(img: Tensor, size: List[int], vertical_flip: bool = False) -> List[
         Corresponding top left, top right, bottom left, bottom right and
         center crop and same for the flipped image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.ten_crop")
+    _log_api_usage_once("transforms", "ten_crop")
     if isinstance(size, numbers.Number):
         size = (int(size), int(size))
     elif isinstance(size, (tuple, list)) and len(size) == 1:
@@ -805,7 +805,7 @@ def adjust_brightness(img: Tensor, brightness_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Brightness adjusted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.adjust_brightness")
+    _log_api_usage_once("transforms", "adjust_brightness")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_brightness(img, brightness_factor)
 
@@ -826,7 +826,7 @@ def adjust_contrast(img: Tensor, contrast_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Contrast adjusted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.adjust_contrast")
+    _log_api_usage_once("transforms", "adjust_contrast")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_contrast(img, contrast_factor)
 
@@ -847,7 +847,7 @@ def adjust_saturation(img: Tensor, saturation_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Saturation adjusted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.adjust_saturation")
+    _log_api_usage_once("transforms", "adjust_saturation")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_saturation(img, saturation_factor)
 
@@ -882,7 +882,7 @@ def adjust_hue(img: Tensor, hue_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Hue adjusted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.adjust_hue")
+    _log_api_usage_once("transforms", "adjust_hue")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_hue(img, hue_factor)
 
@@ -914,7 +914,7 @@ def adjust_gamma(img: Tensor, gamma: float, gain: float = 1) -> Tensor:
     Returns:
         PIL Image or Tensor: Gamma correction adjusted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.adjust_gamma")
+    _log_api_usage_once("transforms", "adjust_gamma")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_gamma(img, gamma, gain)
 
@@ -1011,7 +1011,7 @@ def rotate(
     .. _filters: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#filters
 
     """
-    _log_api_usage_once("torchvision.transforms.functional.rotate")
+    _log_api_usage_once("transforms", "rotate")
     if resample is not None:
         warnings.warn(
             "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1092,7 +1092,7 @@ def affine(
     Returns:
         PIL Image or Tensor: Transformed image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.affine")
+    _log_api_usage_once("transforms", "affine")
     if resample is not None:
         warnings.warn(
             "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1177,7 +1177,7 @@ def to_grayscale(img, num_output_channels=1):
         - if num_output_channels = 1 : returned image is single channel
         - if num_output_channels = 3 : returned image is 3 channel with r = g = b
     """
-    _log_api_usage_once("torchvision.transforms.functional.to_grayscale")
+    _log_api_usage_once("transforms", "to_grayscale")
     if isinstance(img, Image.Image):
         return F_pil.to_grayscale(img, num_output_channels)
 
@@ -1203,7 +1203,7 @@ def rgb_to_grayscale(img: Tensor, num_output_channels: int = 1) -> Tensor:
         - if num_output_channels = 1 : returned image is single channel
         - if num_output_channels = 3 : returned image is 3 channel with r = g = b
     """
-    _log_api_usage_once("torchvision.transforms.functional.rgb_to_grayscale")
+    _log_api_usage_once("transforms", "rgb_to_grayscale")
     if not isinstance(img, torch.Tensor):
         return F_pil.to_grayscale(img, num_output_channels)
 
@@ -1226,7 +1226,7 @@ def erase(img: Tensor, i: int, j: int, h: int, w: int, v: Tensor, inplace: bool
     Returns:
         Tensor Image: Erased image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.erase")
+    _log_api_usage_once("transforms", "erase")
     if not isinstance(img, torch.Tensor):
         raise TypeError(f"img should be Tensor Image. Got {type(img)}")
 
@@ -1263,7 +1263,7 @@ def gaussian_blur(img: Tensor, kernel_size: List[int], sigma: Optional[List[floa
     Returns:
         PIL Image or Tensor: Gaussian Blurred version of the image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.gaussian_blur")
+    _log_api_usage_once("transforms", "gaussian_blur")
     if not isinstance(kernel_size, (int, list, tuple)):
         raise TypeError(f"kernel_size should be int or a sequence of integers. Got {type(kernel_size)}")
     if isinstance(kernel_size, int):
@@ -1315,7 +1315,7 @@ def invert(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: Color inverted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.invert")
+    _log_api_usage_once("transforms", "invert")
     if not isinstance(img, torch.Tensor):
         return F_pil.invert(img)
 
@@ -1335,7 +1335,7 @@ def posterize(img: Tensor, bits: int) -> Tensor:
     Returns:
         PIL Image or Tensor: Posterized image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.posterize")
+    _log_api_usage_once("transforms", "posterize")
     if not (0 <= bits <= 8):
         raise ValueError(f"The number if bits should be between 0 and 8. Got {bits}")
 
@@ -1357,7 +1357,7 @@ def solarize(img: Tensor, threshold: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Solarized image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.solarize")
+    _log_api_usage_once("transforms", "solarize")
     if not isinstance(img, torch.Tensor):
         return F_pil.solarize(img, threshold)
 
@@ -1378,7 +1378,7 @@ def adjust_sharpness(img: Tensor, sharpness_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Sharpness adjusted image.
     """
-    _log_api_usage_once("torchvision.transforms.functional.adjust_sharpness")
+    _log_api_usage_once("transforms", "adjust_sharpness")
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_sharpness(img, sharpness_factor)
 
@@ -1399,7 +1399,7 @@ def autocontrast(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: An image that was autocontrasted.
     """
-    _log_api_usage_once("torchvision.transforms.functional.autocontrast")
+    _log_api_usage_once("transforms", "autocontrast")
     if not isinstance(img, torch.Tensor):
         return F_pil.autocontrast(img)
 
@@ -1421,7 +1421,7 @@ def equalize(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: An image that was equalized.
     """
-    _log_api_usage_once("torchvision.transforms.functional.equalize")
+    _log_api_usage_once("transforms", "equalize")
     if not isinstance(img, torch.Tensor):
         return F_pil.equalize(img)
 
diff --git a/torchvision/transforms/transforms.py b/torchvision/transforms/transforms.py
index 37c9a709b25..c7d72ebed8d 100644
--- a/torchvision/transforms/transforms.py
+++ b/torchvision/transforms/transforms.py
@@ -88,6 +88,7 @@ class Compose:
     """
 
     def __init__(self, transforms):
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.transforms = transforms
 
     def __call__(self, img):
@@ -129,6 +130,7 @@ def __call__(self, pic):
         Returns:
             Tensor: Converted image.
         """
+        _log_api_usage_once("transforms", self.__class__.__name__)
         return F.to_tensor(pic)
 
     def __repr__(self):
@@ -153,6 +155,7 @@ def __call__(self, pic):
         Returns:
             Tensor: Converted image.
         """
+        _log_api_usage_once("transforms", self.__class__.__name__)
         return F.pil_to_tensor(pic)
 
     def __repr__(self):
@@ -180,6 +183,7 @@ class ConvertImageDtype(torch.nn.Module):
 
     def __init__(self, dtype: torch.dtype) -> None:
         super().__init__()
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.dtype = dtype
 
     def forward(self, image):
@@ -205,6 +209,7 @@ class ToPILImage:
     """
 
     def __init__(self, mode=None):
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.mode = mode
 
     def __call__(self, pic):
@@ -246,6 +251,7 @@ class Normalize(torch.nn.Module):
 
     def __init__(self, mean, std, inplace=False):
         super().__init__()
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.mean = mean
         self.std = std
         self.inplace = inplace
@@ -310,6 +316,7 @@ class Resize(torch.nn.Module):
 
     def __init__(self, size, interpolation=InterpolationMode.BILINEAR, max_size=None, antialias=None):
         super().__init__()
+        _log_api_usage_once("transforms", self.__class__.__name__)
         if not isinstance(size, (int, Sequence)):
             raise TypeError(f"Size should be int or sequence. Got {type(size)}")
         if isinstance(size, Sequence) and len(size) not in (1, 2):
@@ -351,6 +358,7 @@ class Scale(Resize):
     def __init__(self, *args, **kwargs):
         warnings.warn("The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.")
         super().__init__(*args, **kwargs)
+        _log_api_usage_once("transforms", self.__class__.__name__)
 
 
 class CenterCrop(torch.nn.Module):
@@ -367,6 +375,7 @@ class CenterCrop(torch.nn.Module):
 
     def __init__(self, size):
         super().__init__()
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
 
     def forward(self, img):
@@ -423,6 +432,7 @@ class Pad(torch.nn.Module):
 
     def __init__(self, padding, fill=0, padding_mode="constant"):
         super().__init__()
+        _log_api_usage_once("transforms", self.__class__.__name__)
         if not isinstance(padding, (numbers.Number, tuple, list)):
             raise TypeError("Got inappropriate padding arg")
 
@@ -463,7 +473,7 @@ class Lambda:
     """
 
     def __init__(self, lambd):
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         if not callable(lambd):
             raise TypeError(f"Argument lambd should be callable, got {repr(type(lambd).__name__)}")
         self.lambd = lambd
@@ -483,7 +493,7 @@ class RandomTransforms:
     """
 
     def __init__(self, transforms):
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         if not isinstance(transforms, Sequence):
             raise TypeError("Argument transforms should be a sequence")
         self.transforms = transforms
@@ -522,7 +532,7 @@ class RandomApply(torch.nn.Module):
 
     def __init__(self, transforms, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.transforms = transforms
         self.p = p
 
@@ -643,7 +653,7 @@ def get_params(img: Tensor, output_size: Tuple[int, int]) -> Tuple[int, int, int
 
     def __init__(self, size, padding=None, pad_if_needed=False, fill=0, padding_mode="constant"):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
 
         self.size = tuple(_setup_size(size, error_msg="Please provide only two dimensions (h, w) for size."))
 
@@ -693,7 +703,7 @@ class RandomHorizontalFlip(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.p = p
 
     def forward(self, img):
@@ -724,7 +734,7 @@ class RandomVerticalFlip(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.p = p
 
     def forward(self, img):
@@ -762,7 +772,7 @@ class RandomPerspective(torch.nn.Module):
 
     def __init__(self, distortion_scale=0.5, p=0.5, interpolation=InterpolationMode.BILINEAR, fill=0):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.p = p
 
         # Backward compatibility with integer value
@@ -875,7 +885,7 @@ class RandomResizedCrop(torch.nn.Module):
 
     def __init__(self, size, scale=(0.08, 1.0), ratio=(3.0 / 4.0, 4.0 / 3.0), interpolation=InterpolationMode.BILINEAR):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
 
         if not isinstance(scale, Sequence):
@@ -972,6 +982,7 @@ def __init__(self, *args, **kwargs):
             + "please use transforms.RandomResizedCrop instead."
         )
         super().__init__(*args, **kwargs)
+        _log_api_usage_once("transforms", self.__class__.__name__)
 
 
 class FiveCrop(torch.nn.Module):
@@ -1004,6 +1015,7 @@ class FiveCrop(torch.nn.Module):
 
     def __init__(self, size):
         super().__init__()
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
 
     def forward(self, img):
@@ -1052,6 +1064,7 @@ class TenCrop(torch.nn.Module):
 
     def __init__(self, size, vertical_flip=False):
         super().__init__()
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
         self.vertical_flip = vertical_flip
 
@@ -1090,7 +1103,7 @@ class LinearTransformation(torch.nn.Module):
 
     def __init__(self, transformation_matrix, mean_vector):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         if transformation_matrix.size(0) != transformation_matrix.size(1):
             raise ValueError(
                 "transformation_matrix should be square. Got "
@@ -1169,7 +1182,7 @@ class ColorJitter(torch.nn.Module):
 
     def __init__(self, brightness=0, contrast=0, saturation=0, hue=0):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.brightness = self._check_input(brightness, "brightness")
         self.contrast = self._check_input(contrast, "contrast")
         self.saturation = self._check_input(saturation, "saturation")
@@ -1292,7 +1305,7 @@ def __init__(
         self, degrees, interpolation=InterpolationMode.NEAREST, expand=False, center=None, fill=0, resample=None
     ):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         if resample is not None:
             warnings.warn(
                 "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1413,7 +1426,7 @@ def __init__(
         resample=None,
     ):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         if resample is not None:
             warnings.warn(
                 "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1558,7 +1571,7 @@ class Grayscale(torch.nn.Module):
 
     def __init__(self, num_output_channels=1):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.num_output_channels = num_output_channels
 
     def forward(self, img):
@@ -1593,7 +1606,7 @@ class RandomGrayscale(torch.nn.Module):
 
     def __init__(self, p=0.1):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.p = p
 
     def forward(self, img):
@@ -1643,7 +1656,7 @@ class RandomErasing(torch.nn.Module):
 
     def __init__(self, p=0.5, scale=(0.02, 0.33), ratio=(0.3, 3.3), value=0, inplace=False):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         if not isinstance(value, (numbers.Number, str, tuple, list)):
             raise TypeError("Argument value should be either a number or str or a sequence")
         if isinstance(value, str) and value != "random":
@@ -1767,7 +1780,7 @@ class GaussianBlur(torch.nn.Module):
 
     def __init__(self, kernel_size, sigma=(0.1, 2.0)):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.kernel_size = _setup_size(kernel_size, "Kernel size should be a tuple/list of two integers")
         for ks in self.kernel_size:
             if ks <= 0 or ks % 2 == 0:
@@ -1859,7 +1872,7 @@ class RandomInvert(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.p = p
 
     def forward(self, img):
@@ -1891,7 +1904,7 @@ class RandomPosterize(torch.nn.Module):
 
     def __init__(self, bits, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.bits = bits
         self.p = p
 
@@ -1924,7 +1937,7 @@ class RandomSolarize(torch.nn.Module):
 
     def __init__(self, threshold, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.threshold = threshold
         self.p = p
 
@@ -1957,7 +1970,7 @@ class RandomAdjustSharpness(torch.nn.Module):
 
     def __init__(self, sharpness_factor, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.sharpness_factor = sharpness_factor
         self.p = p
 
@@ -1989,7 +2002,7 @@ class RandomAutocontrast(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.p = p
 
     def forward(self, img):
@@ -2020,7 +2033,7 @@ class RandomEqualize(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once(self)
+        _log_api_usage_once("transforms", self.__class__.__name__)
         self.p = p
 
     def forward(self, img):

From d0b9bf5af6b37a483a6ca5266a8d5d693368f76d Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Fri, 10 Dec 2021 07:21:33 +0000
Subject: [PATCH 6/9] torchscriptbility

---
 torchvision/transforms/transforms.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/torchvision/transforms/transforms.py b/torchvision/transforms/transforms.py
index c7d72ebed8d..371b043f9a0 100644
--- a/torchvision/transforms/transforms.py
+++ b/torchvision/transforms/transforms.py
@@ -88,7 +88,7 @@ class Compose:
     """
 
     def __init__(self, transforms):
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once("transforms", "Compose")
         self.transforms = transforms
 
     def __call__(self, img):

From b209d85231767bf09a1cf662f42fcdeed885c80c Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Mon, 20 Dec 2021 19:35:52 +0000
Subject: [PATCH 7/9] adopt new API

---
 torchvision/transforms/functional.py | 105 ++++++++++++++++++---------
 torchvision/transforms/transforms.py |  70 +++++++++---------
 2 files changed, 105 insertions(+), 70 deletions(-)

diff --git a/torchvision/transforms/functional.py b/torchvision/transforms/functional.py
index 99f4ce7e203..7d7d5382291 100644
--- a/torchvision/transforms/functional.py
+++ b/torchvision/transforms/functional.py
@@ -68,7 +68,8 @@ def get_image_size(img: Tensor) -> List[int]:
     Returns:
         List[int]: The image size.
     """
-    _log_api_usage_once("transforms", "get_image_size")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(get_image_size)
     if isinstance(img, torch.Tensor):
         return F_t.get_image_size(img)
 
@@ -84,7 +85,8 @@ def get_image_num_channels(img: Tensor) -> int:
     Returns:
         int: The number of channels.
     """
-    _log_api_usage_once("transforms", "get_image_num_channels")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(get_image_num_channels)
     if isinstance(img, torch.Tensor):
         return F_t.get_image_num_channels(img)
 
@@ -113,7 +115,8 @@ def to_tensor(pic):
     Returns:
         Tensor: Converted image.
     """
-    _log_api_usage_once("transforms", "to_tensor")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(to_tensor)
     if not (F_pil._is_pil_image(pic) or _is_numpy(pic)):
         raise TypeError(f"pic should be PIL Image or ndarray. Got {type(pic)}")
 
@@ -170,7 +173,8 @@ def pil_to_tensor(pic):
     Returns:
         Tensor: Converted image.
     """
-    _log_api_usage_once("transforms", "pil_to_tensor")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(pil_to_tensor)
     if not F_pil._is_pil_image(pic):
         raise TypeError(f"pic should be PIL Image. Got {type(pic)}")
 
@@ -210,7 +214,8 @@ def convert_image_dtype(image: torch.Tensor, dtype: torch.dtype = torch.float) -
             overflow errors since the floating point ``dtype`` cannot store consecutive integers over the whole range
             of the integer ``dtype``.
     """
-    _log_api_usage_once("transforms", "convert_image_dtype")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(convert_image_dtype)
     if not isinstance(image, torch.Tensor):
         raise TypeError("Input img should be Tensor Image")
 
@@ -231,7 +236,8 @@ def to_pil_image(pic, mode=None):
     Returns:
         PIL Image: Image converted to PIL Image.
     """
-    _log_api_usage_once("transforms", "to_pil_image")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(to_pil_image)
     if not (isinstance(pic, torch.Tensor) or isinstance(pic, np.ndarray)):
         raise TypeError(f"pic should be Tensor or ndarray. Got {type(pic)}.")
 
@@ -329,7 +335,8 @@ def normalize(tensor: Tensor, mean: List[float], std: List[float], inplace: bool
     Returns:
         Tensor: Normalized Tensor image.
     """
-    _log_api_usage_once("transforms", "normalize")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(normalize)
     if not isinstance(tensor, torch.Tensor):
         raise TypeError(f"Input tensor should be a torch tensor. Got {type(tensor)}.")
 
@@ -409,7 +416,8 @@ def resize(
     Returns:
         PIL Image or Tensor: Resized image.
     """
-    _log_api_usage_once("transforms", "resize")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(resize)
     # Backward compatibility with integer value
     if isinstance(interpolation, int):
         warnings.warn(
@@ -431,7 +439,8 @@ def resize(
 
 
 def scale(*args, **kwargs):
-    _log_api_usage_once("transforms", "scale")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(scale)
     warnings.warn("The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.")
     return resize(*args, **kwargs)
 
@@ -477,7 +486,8 @@ def pad(img: Tensor, padding: List[int], fill: int = 0, padding_mode: str = "con
     Returns:
         PIL Image or Tensor: Padded image.
     """
-    _log_api_usage_once("transforms", "pad")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(pad)
     if not isinstance(img, torch.Tensor):
         return F_pil.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
 
@@ -501,7 +511,8 @@ def crop(img: Tensor, top: int, left: int, height: int, width: int) -> Tensor:
         PIL Image or Tensor: Cropped image.
     """
 
-    _log_api_usage_once("transforms", "crop")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(crop)
     if not isinstance(img, torch.Tensor):
         return F_pil.crop(img, top, left, height, width)
 
@@ -522,7 +533,8 @@ def center_crop(img: Tensor, output_size: List[int]) -> Tensor:
     Returns:
         PIL Image or Tensor: Cropped image.
     """
-    _log_api_usage_once("transforms", "center_crop")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(center_crop)
     if isinstance(output_size, numbers.Number):
         output_size = (int(output_size), int(output_size))
     elif isinstance(output_size, (tuple, list)) and len(output_size) == 1:
@@ -579,7 +591,8 @@ def resized_crop(
     Returns:
         PIL Image or Tensor: Cropped image.
     """
-    _log_api_usage_once("transforms", "resized_crop")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(resized_crop)
     img = crop(img, top, left, height, width)
     img = resize(img, size, interpolation)
     return img
@@ -597,7 +610,8 @@ def hflip(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor:  Horizontally flipped image.
     """
-    _log_api_usage_once("transforms", "hflip")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(hflip)
     if not isinstance(img, torch.Tensor):
         return F_pil.hflip(img)
 
@@ -663,7 +677,8 @@ def perspective(
     Returns:
         PIL Image or Tensor: transformed Image.
     """
-    _log_api_usage_once("transforms", "perspective")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(perspective)
 
     coeffs = _get_perspective_coeffs(startpoints, endpoints)
 
@@ -697,7 +712,8 @@ def vflip(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor:  Vertically flipped image.
     """
-    _log_api_usage_once("transforms", "vflip")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(vflip)
     if not isinstance(img, torch.Tensor):
         return F_pil.vflip(img)
 
@@ -723,7 +739,8 @@ def five_crop(img: Tensor, size: List[int]) -> Tuple[Tensor, Tensor, Tensor, Ten
        tuple: tuple (tl, tr, bl, br, center)
        Corresponding top left, top right, bottom left, bottom right and center crop.
     """
-    _log_api_usage_once("transforms", "five_crop")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(five_crop)
     if isinstance(size, numbers.Number):
         size = (int(size), int(size))
     elif isinstance(size, (tuple, list)) and len(size) == 1:
@@ -771,7 +788,8 @@ def ten_crop(img: Tensor, size: List[int], vertical_flip: bool = False) -> List[
         Corresponding top left, top right, bottom left, bottom right and
         center crop and same for the flipped image.
     """
-    _log_api_usage_once("transforms", "ten_crop")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(ten_crop)
     if isinstance(size, numbers.Number):
         size = (int(size), int(size))
     elif isinstance(size, (tuple, list)) and len(size) == 1:
@@ -805,7 +823,8 @@ def adjust_brightness(img: Tensor, brightness_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Brightness adjusted image.
     """
-    _log_api_usage_once("transforms", "adjust_brightness")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(adjust_brightness)
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_brightness(img, brightness_factor)
 
@@ -826,7 +845,8 @@ def adjust_contrast(img: Tensor, contrast_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Contrast adjusted image.
     """
-    _log_api_usage_once("transforms", "adjust_contrast")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(adjust_contrast)
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_contrast(img, contrast_factor)
 
@@ -847,7 +867,8 @@ def adjust_saturation(img: Tensor, saturation_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Saturation adjusted image.
     """
-    _log_api_usage_once("transforms", "adjust_saturation")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(adjust_saturation)
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_saturation(img, saturation_factor)
 
@@ -882,7 +903,8 @@ def adjust_hue(img: Tensor, hue_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Hue adjusted image.
     """
-    _log_api_usage_once("transforms", "adjust_hue")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(adjust_hue)
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_hue(img, hue_factor)
 
@@ -914,7 +936,8 @@ def adjust_gamma(img: Tensor, gamma: float, gain: float = 1) -> Tensor:
     Returns:
         PIL Image or Tensor: Gamma correction adjusted image.
     """
-    _log_api_usage_once("transforms", "adjust_gamma")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(adjust_gamma)
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_gamma(img, gamma, gain)
 
@@ -1011,7 +1034,8 @@ def rotate(
     .. _filters: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#filters
 
     """
-    _log_api_usage_once("transforms", "rotate")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(rotate)
     if resample is not None:
         warnings.warn(
             "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1092,7 +1116,8 @@ def affine(
     Returns:
         PIL Image or Tensor: Transformed image.
     """
-    _log_api_usage_once("transforms", "affine")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(affine)
     if resample is not None:
         warnings.warn(
             "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1177,7 +1202,8 @@ def to_grayscale(img, num_output_channels=1):
         - if num_output_channels = 1 : returned image is single channel
         - if num_output_channels = 3 : returned image is 3 channel with r = g = b
     """
-    _log_api_usage_once("transforms", "to_grayscale")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(to_grayscale)
     if isinstance(img, Image.Image):
         return F_pil.to_grayscale(img, num_output_channels)
 
@@ -1203,7 +1229,8 @@ def rgb_to_grayscale(img: Tensor, num_output_channels: int = 1) -> Tensor:
         - if num_output_channels = 1 : returned image is single channel
         - if num_output_channels = 3 : returned image is 3 channel with r = g = b
     """
-    _log_api_usage_once("transforms", "rgb_to_grayscale")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(rgb_to_grayscale)
     if not isinstance(img, torch.Tensor):
         return F_pil.to_grayscale(img, num_output_channels)
 
@@ -1226,7 +1253,8 @@ def erase(img: Tensor, i: int, j: int, h: int, w: int, v: Tensor, inplace: bool
     Returns:
         Tensor Image: Erased image.
     """
-    _log_api_usage_once("transforms", "erase")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(erase)
     if not isinstance(img, torch.Tensor):
         raise TypeError(f"img should be Tensor Image. Got {type(img)}")
 
@@ -1263,7 +1291,8 @@ def gaussian_blur(img: Tensor, kernel_size: List[int], sigma: Optional[List[floa
     Returns:
         PIL Image or Tensor: Gaussian Blurred version of the image.
     """
-    _log_api_usage_once("transforms", "gaussian_blur")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(gaussian_blur)
     if not isinstance(kernel_size, (int, list, tuple)):
         raise TypeError(f"kernel_size should be int or a sequence of integers. Got {type(kernel_size)}")
     if isinstance(kernel_size, int):
@@ -1315,7 +1344,8 @@ def invert(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: Color inverted image.
     """
-    _log_api_usage_once("transforms", "invert")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(invert)
     if not isinstance(img, torch.Tensor):
         return F_pil.invert(img)
 
@@ -1335,7 +1365,8 @@ def posterize(img: Tensor, bits: int) -> Tensor:
     Returns:
         PIL Image or Tensor: Posterized image.
     """
-    _log_api_usage_once("transforms", "posterize")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(posterize)
     if not (0 <= bits <= 8):
         raise ValueError(f"The number if bits should be between 0 and 8. Got {bits}")
 
@@ -1357,7 +1388,8 @@ def solarize(img: Tensor, threshold: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Solarized image.
     """
-    _log_api_usage_once("transforms", "solarize")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(solarize)
     if not isinstance(img, torch.Tensor):
         return F_pil.solarize(img, threshold)
 
@@ -1378,7 +1410,8 @@ def adjust_sharpness(img: Tensor, sharpness_factor: float) -> Tensor:
     Returns:
         PIL Image or Tensor: Sharpness adjusted image.
     """
-    _log_api_usage_once("transforms", "adjust_sharpness")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(adjust_sharpness)
     if not isinstance(img, torch.Tensor):
         return F_pil.adjust_sharpness(img, sharpness_factor)
 
@@ -1399,7 +1432,8 @@ def autocontrast(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: An image that was autocontrasted.
     """
-    _log_api_usage_once("transforms", "autocontrast")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(autocontrast)
     if not isinstance(img, torch.Tensor):
         return F_pil.autocontrast(img)
 
@@ -1421,7 +1455,8 @@ def equalize(img: Tensor) -> Tensor:
     Returns:
         PIL Image or Tensor: An image that was equalized.
     """
-    _log_api_usage_once("transforms", "equalize")
+    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+        _log_api_usage_once(equalize)
     if not isinstance(img, torch.Tensor):
         return F_pil.equalize(img)
 
diff --git a/torchvision/transforms/transforms.py b/torchvision/transforms/transforms.py
index 371b043f9a0..cd988d410c4 100644
--- a/torchvision/transforms/transforms.py
+++ b/torchvision/transforms/transforms.py
@@ -88,7 +88,7 @@ class Compose:
     """
 
     def __init__(self, transforms):
-        _log_api_usage_once("transforms", "Compose")
+        _log_api_usage_once(self)
         self.transforms = transforms
 
     def __call__(self, img):
@@ -130,7 +130,7 @@ def __call__(self, pic):
         Returns:
             Tensor: Converted image.
         """
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         return F.to_tensor(pic)
 
     def __repr__(self):
@@ -155,7 +155,7 @@ def __call__(self, pic):
         Returns:
             Tensor: Converted image.
         """
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         return F.pil_to_tensor(pic)
 
     def __repr__(self):
@@ -183,7 +183,7 @@ class ConvertImageDtype(torch.nn.Module):
 
     def __init__(self, dtype: torch.dtype) -> None:
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.dtype = dtype
 
     def forward(self, image):
@@ -209,7 +209,7 @@ class ToPILImage:
     """
 
     def __init__(self, mode=None):
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.mode = mode
 
     def __call__(self, pic):
@@ -251,7 +251,7 @@ class Normalize(torch.nn.Module):
 
     def __init__(self, mean, std, inplace=False):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.mean = mean
         self.std = std
         self.inplace = inplace
@@ -316,7 +316,7 @@ class Resize(torch.nn.Module):
 
     def __init__(self, size, interpolation=InterpolationMode.BILINEAR, max_size=None, antialias=None):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         if not isinstance(size, (int, Sequence)):
             raise TypeError(f"Size should be int or sequence. Got {type(size)}")
         if isinstance(size, Sequence) and len(size) not in (1, 2):
@@ -358,7 +358,7 @@ class Scale(Resize):
     def __init__(self, *args, **kwargs):
         warnings.warn("The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.")
         super().__init__(*args, **kwargs)
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
 
 
 class CenterCrop(torch.nn.Module):
@@ -375,7 +375,7 @@ class CenterCrop(torch.nn.Module):
 
     def __init__(self, size):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
 
     def forward(self, img):
@@ -432,7 +432,7 @@ class Pad(torch.nn.Module):
 
     def __init__(self, padding, fill=0, padding_mode="constant"):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         if not isinstance(padding, (numbers.Number, tuple, list)):
             raise TypeError("Got inappropriate padding arg")
 
@@ -473,7 +473,7 @@ class Lambda:
     """
 
     def __init__(self, lambd):
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         if not callable(lambd):
             raise TypeError(f"Argument lambd should be callable, got {repr(type(lambd).__name__)}")
         self.lambd = lambd
@@ -493,7 +493,7 @@ class RandomTransforms:
     """
 
     def __init__(self, transforms):
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         if not isinstance(transforms, Sequence):
             raise TypeError("Argument transforms should be a sequence")
         self.transforms = transforms
@@ -532,7 +532,7 @@ class RandomApply(torch.nn.Module):
 
     def __init__(self, transforms, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.transforms = transforms
         self.p = p
 
@@ -653,7 +653,7 @@ def get_params(img: Tensor, output_size: Tuple[int, int]) -> Tuple[int, int, int
 
     def __init__(self, size, padding=None, pad_if_needed=False, fill=0, padding_mode="constant"):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
 
         self.size = tuple(_setup_size(size, error_msg="Please provide only two dimensions (h, w) for size."))
 
@@ -703,7 +703,7 @@ class RandomHorizontalFlip(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -734,7 +734,7 @@ class RandomVerticalFlip(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -772,7 +772,7 @@ class RandomPerspective(torch.nn.Module):
 
     def __init__(self, distortion_scale=0.5, p=0.5, interpolation=InterpolationMode.BILINEAR, fill=0):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.p = p
 
         # Backward compatibility with integer value
@@ -885,7 +885,7 @@ class RandomResizedCrop(torch.nn.Module):
 
     def __init__(self, size, scale=(0.08, 1.0), ratio=(3.0 / 4.0, 4.0 / 3.0), interpolation=InterpolationMode.BILINEAR):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
 
         if not isinstance(scale, Sequence):
@@ -982,7 +982,7 @@ def __init__(self, *args, **kwargs):
             + "please use transforms.RandomResizedCrop instead."
         )
         super().__init__(*args, **kwargs)
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
 
 
 class FiveCrop(torch.nn.Module):
@@ -1015,7 +1015,7 @@ class FiveCrop(torch.nn.Module):
 
     def __init__(self, size):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
 
     def forward(self, img):
@@ -1064,7 +1064,7 @@ class TenCrop(torch.nn.Module):
 
     def __init__(self, size, vertical_flip=False):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.size = _setup_size(size, error_msg="Please provide only two dimensions (h, w) for size.")
         self.vertical_flip = vertical_flip
 
@@ -1103,7 +1103,7 @@ class LinearTransformation(torch.nn.Module):
 
     def __init__(self, transformation_matrix, mean_vector):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         if transformation_matrix.size(0) != transformation_matrix.size(1):
             raise ValueError(
                 "transformation_matrix should be square. Got "
@@ -1182,7 +1182,7 @@ class ColorJitter(torch.nn.Module):
 
     def __init__(self, brightness=0, contrast=0, saturation=0, hue=0):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.brightness = self._check_input(brightness, "brightness")
         self.contrast = self._check_input(contrast, "contrast")
         self.saturation = self._check_input(saturation, "saturation")
@@ -1305,7 +1305,7 @@ def __init__(
         self, degrees, interpolation=InterpolationMode.NEAREST, expand=False, center=None, fill=0, resample=None
     ):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         if resample is not None:
             warnings.warn(
                 "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1426,7 +1426,7 @@ def __init__(
         resample=None,
     ):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         if resample is not None:
             warnings.warn(
                 "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
@@ -1571,7 +1571,7 @@ class Grayscale(torch.nn.Module):
 
     def __init__(self, num_output_channels=1):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.num_output_channels = num_output_channels
 
     def forward(self, img):
@@ -1606,7 +1606,7 @@ class RandomGrayscale(torch.nn.Module):
 
     def __init__(self, p=0.1):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -1656,7 +1656,7 @@ class RandomErasing(torch.nn.Module):
 
     def __init__(self, p=0.5, scale=(0.02, 0.33), ratio=(0.3, 3.3), value=0, inplace=False):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         if not isinstance(value, (numbers.Number, str, tuple, list)):
             raise TypeError("Argument value should be either a number or str or a sequence")
         if isinstance(value, str) and value != "random":
@@ -1780,7 +1780,7 @@ class GaussianBlur(torch.nn.Module):
 
     def __init__(self, kernel_size, sigma=(0.1, 2.0)):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.kernel_size = _setup_size(kernel_size, "Kernel size should be a tuple/list of two integers")
         for ks in self.kernel_size:
             if ks <= 0 or ks % 2 == 0:
@@ -1872,7 +1872,7 @@ class RandomInvert(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -1904,7 +1904,7 @@ class RandomPosterize(torch.nn.Module):
 
     def __init__(self, bits, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.bits = bits
         self.p = p
 
@@ -1937,7 +1937,7 @@ class RandomSolarize(torch.nn.Module):
 
     def __init__(self, threshold, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.threshold = threshold
         self.p = p
 
@@ -1970,7 +1970,7 @@ class RandomAdjustSharpness(torch.nn.Module):
 
     def __init__(self, sharpness_factor, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.sharpness_factor = sharpness_factor
         self.p = p
 
@@ -2002,7 +2002,7 @@ class RandomAutocontrast(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):
@@ -2033,7 +2033,7 @@ class RandomEqualize(torch.nn.Module):
 
     def __init__(self, p=0.5):
         super().__init__()
-        _log_api_usage_once("transforms", self.__class__.__name__)
+        _log_api_usage_once(self)
         self.p = p
 
     def forward(self, img):

From 06a93c357169d5e9290f0ef604c836b83ee70d69 Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Mon, 20 Dec 2021 22:15:43 +0000
Subject: [PATCH 8/9] make Compose scriptable

---
 torchvision/transforms/transforms.py | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)

diff --git a/torchvision/transforms/transforms.py b/torchvision/transforms/transforms.py
index cd988d410c4..fc7b1bd8ba8 100644
--- a/torchvision/transforms/transforms.py
+++ b/torchvision/transforms/transforms.py
@@ -88,7 +88,8 @@ class Compose:
     """
 
     def __init__(self, transforms):
-        _log_api_usage_once(self)
+        if not torch.jit.is_scripting() and not torch.jit.is_tracing():
+            _log_api_usage_once(self)
         self.transforms = transforms
 
     def __call__(self, img):

From a84e56ad37b000f581da2e41d0cd08550e08530a Mon Sep 17 00:00:00 2001
From: Kai Zhang <kaizh@fb.com>
Date: Tue, 21 Dec 2021 01:28:42 +0000
Subject: [PATCH 9/9] move from __call__ to __init__

---
 torchvision/transforms/transforms.py | 8 ++++++--
 1 file changed, 6 insertions(+), 2 deletions(-)

diff --git a/torchvision/transforms/transforms.py b/torchvision/transforms/transforms.py
index fc7b1bd8ba8..5cacecab625 100644
--- a/torchvision/transforms/transforms.py
+++ b/torchvision/transforms/transforms.py
@@ -123,6 +123,9 @@ class ToTensor:
     .. _references: https://github.com/pytorch/vision/tree/main/references/segmentation
     """
 
+    def __init__(self) -> None:
+        _log_api_usage_once(self)
+
     def __call__(self, pic):
         """
         Args:
@@ -131,7 +134,6 @@ def __call__(self, pic):
         Returns:
             Tensor: Converted image.
         """
-        _log_api_usage_once(self)
         return F.to_tensor(pic)
 
     def __repr__(self):
@@ -144,6 +146,9 @@ class PILToTensor:
     Converts a PIL Image (H x W x C) to a Tensor of shape (C x H x W).
     """
 
+    def __init__(self) -> None:
+        _log_api_usage_once(self)
+
     def __call__(self, pic):
         """
         .. note::
@@ -156,7 +161,6 @@ def __call__(self, pic):
         Returns:
             Tensor: Converted image.
         """
-        _log_api_usage_once(self)
         return F.pil_to_tensor(pic)
 
     def __repr__(self):