functional.py 64.6 KB
Newer Older
1
import math
2
3
import numbers
import warnings
4
from enum import Enum
5
from typing import List, Tuple, Any, Optional
6
7
8

import numpy as np
import torch
9
from PIL import Image
10
11
from torch import Tensor

12
13
14
15
16
try:
    import accimage
except ImportError:
    accimage = None

17
from ..utils import _log_api_usage_once
18
19
20
from . import functional_pil as F_pil
from . import functional_tensor as F_t

21

22
class InterpolationMode(Enum):
23
    """Interpolation modes
24
    Available interpolation methods are ``nearest``, ``bilinear``, ``bicubic``, ``box``, ``hamming``, and ``lanczos``.
25
    """
26

27
28
29
30
31
32
33
34
35
36
    NEAREST = "nearest"
    BILINEAR = "bilinear"
    BICUBIC = "bicubic"
    # For PIL compatibility
    BOX = "box"
    HAMMING = "hamming"
    LANCZOS = "lanczos"


# TODO: Once torchscript supports Enums with staticmethod
37
38
# this can be put into InterpolationMode as staticmethod
def _interpolation_modes_from_int(i: int) -> InterpolationMode:
39
    inverse_modes_mapping = {
40
41
42
43
44
45
        0: InterpolationMode.NEAREST,
        2: InterpolationMode.BILINEAR,
        3: InterpolationMode.BICUBIC,
        4: InterpolationMode.BOX,
        5: InterpolationMode.HAMMING,
        1: InterpolationMode.LANCZOS,
46
47
48
49
50
    }
    return inverse_modes_mapping[i]


pil_modes_mapping = {
51
52
53
54
55
56
    InterpolationMode.NEAREST: 0,
    InterpolationMode.BILINEAR: 2,
    InterpolationMode.BICUBIC: 3,
    InterpolationMode.BOX: 4,
    InterpolationMode.HAMMING: 5,
    InterpolationMode.LANCZOS: 1,
57
58
}

vfdev's avatar
vfdev committed
59
60
61
_is_pil_image = F_pil._is_pil_image


62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
def get_dimensions(img: Tensor) -> List[int]:
    """Returns the dimensions of an image as [channels, height, width].

    Args:
        img (PIL Image or Tensor): The image to be checked.

    Returns:
        List[int]: The image dimensions.
    """
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(get_dimensions)
    if isinstance(img, torch.Tensor):
        return F_t.get_dimensions(img)

    return F_pil.get_dimensions(img)


79
80
81
82
83
84
85
86
def get_image_size(img: Tensor) -> List[int]:
    """Returns the size of an image as [width, height].

    Args:
        img (PIL Image or Tensor): The image to be checked.

    Returns:
        List[int]: The image size.
vfdev's avatar
vfdev committed
87
    """
88
89
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(get_image_size)
vfdev's avatar
vfdev committed
90
    if isinstance(img, torch.Tensor):
91
        return F_t.get_image_size(img)
92

93
    return F_pil.get_image_size(img)
94

vfdev's avatar
vfdev committed
95

96
97
98
99
100
101
102
103
def get_image_num_channels(img: Tensor) -> int:
    """Returns the number of channels of an image.

    Args:
        img (PIL Image or Tensor): The image to be checked.

    Returns:
        int: The number of channels.
104
    """
105
106
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(get_image_num_channels)
107
    if isinstance(img, torch.Tensor):
108
        return F_t.get_image_num_channels(img)
109

110
    return F_pil.get_image_num_channels(img)
111
112


vfdev's avatar
vfdev committed
113
114
@torch.jit.unused
def _is_numpy(img: Any) -> bool:
115
116
117
    return isinstance(img, np.ndarray)


vfdev's avatar
vfdev committed
118
119
@torch.jit.unused
def _is_numpy_image(img: Any) -> bool:
120
    return img.ndim in {2, 3}
121
122


123
def to_tensor(pic) -> Tensor:
124
    """Convert a ``PIL Image`` or ``numpy.ndarray`` to tensor.
125
    This function does not support torchscript.
126

127
    See :class:`~torchvision.transforms.ToTensor` for more details.
128
129
130
131
132
133
134

    Args:
        pic (PIL Image or numpy.ndarray): Image to be converted to tensor.

    Returns:
        Tensor: Converted image.
    """
135
136
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(to_tensor)
137
    if not (F_pil._is_pil_image(pic) or _is_numpy(pic)):
138
        raise TypeError(f"pic should be PIL Image or ndarray. Got {type(pic)}")
139

140
    if _is_numpy(pic) and not _is_numpy_image(pic):
141
        raise ValueError(f"pic should be 2/3 dimensional. Got {pic.ndim} dimensions.")
142

143
144
    default_float_dtype = torch.get_default_dtype()

145
146
    if isinstance(pic, np.ndarray):
        # handle numpy array
surgan12's avatar
surgan12 committed
147
148
149
        if pic.ndim == 2:
            pic = pic[:, :, None]

150
        img = torch.from_numpy(pic.transpose((2, 0, 1))).contiguous()
151
        # backward compatibility
152
        if isinstance(img, torch.ByteTensor):
153
            return img.to(dtype=default_float_dtype).div(255)
154
155
        else:
            return img
156
157

    if accimage is not None and isinstance(pic, accimage.Image):
158
        nppic = np.zeros([pic.channels, pic.height, pic.width], dtype=np.float32)
159
        pic.copyto(nppic)
160
        return torch.from_numpy(nppic).to(dtype=default_float_dtype)
161
162

    # handle PIL Image
163
164
    mode_to_nptype = {"I": np.int32, "I;16": np.int16, "F": np.float32}
    img = torch.from_numpy(np.array(pic, mode_to_nptype.get(pic.mode, np.uint8), copy=True))
165

166
    if pic.mode == "1":
167
        img = 255 * img
168
    img = img.view(pic.size[1], pic.size[0], len(pic.getbands()))
169
    # put it from HWC to CHW format
170
    img = img.permute((2, 0, 1)).contiguous()
171
    if isinstance(img, torch.ByteTensor):
172
        return img.to(dtype=default_float_dtype).div(255)
173
174
175
176
    else:
        return img


177
178
def pil_to_tensor(pic):
    """Convert a ``PIL Image`` to a tensor of the same type.
179
    This function does not support torchscript.
180

vfdev's avatar
vfdev committed
181
    See :class:`~torchvision.transforms.PILToTensor` for more details.
182

183
184
185
186
    .. note::

        A deep copy of the underlying array is performed.

187
188
189
190
191
192
    Args:
        pic (PIL Image): Image to be converted to tensor.

    Returns:
        Tensor: Converted image.
    """
193
194
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(pil_to_tensor)
195
    if not F_pil._is_pil_image(pic):
196
        raise TypeError(f"pic should be PIL Image. Got {type(pic)}")
197
198

    if accimage is not None and isinstance(pic, accimage.Image):
199
200
        # accimage format is always uint8 internally, so always return uint8 here
        nppic = np.zeros([pic.channels, pic.height, pic.width], dtype=np.uint8)
201
202
203
204
        pic.copyto(nppic)
        return torch.as_tensor(nppic)

    # handle PIL Image
205
    img = torch.as_tensor(np.array(pic, copy=True))
206
207
208
209
210
211
    img = img.view(pic.size[1], pic.size[0], len(pic.getbands()))
    # put it from HWC to CHW format
    img = img.permute((2, 0, 1))
    return img


212
213
def convert_image_dtype(image: torch.Tensor, dtype: torch.dtype = torch.float) -> torch.Tensor:
    """Convert a tensor image to the given ``dtype`` and scale the values accordingly
214
    This function does not support PIL Image.
215
216
217
218
219
220

    Args:
        image (torch.Tensor): Image to be converted
        dtype (torch.dtype): Desired data type of the output

    Returns:
vfdev's avatar
vfdev committed
221
        Tensor: Converted image
222
223
224
225
226
227
228
229
230
231
232
233

    .. note::

        When converting from a smaller to a larger integer ``dtype`` the maximum values are **not** mapped exactly.
        If converted back and forth, this mismatch has no effect.

    Raises:
        RuntimeError: When trying to cast :class:`torch.float32` to :class:`torch.int32` or :class:`torch.int64` as
            well as for trying to cast :class:`torch.float64` to :class:`torch.int64`. These conversions might lead to
            overflow errors since the floating point ``dtype`` cannot store consecutive integers over the whole range
            of the integer ``dtype``.
    """
234
235
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(convert_image_dtype)
236
    if not isinstance(image, torch.Tensor):
237
        raise TypeError("Input img should be Tensor Image")
238
239

    return F_t.convert_image_dtype(image, dtype)
240
241


242
def to_pil_image(pic, mode=None):
243
    """Convert a tensor or an ndarray to PIL Image. This function does not support torchscript.
244

245
    See :class:`~torchvision.transforms.ToPILImage` for more details.
246
247
248
249
250

    Args:
        pic (Tensor or numpy.ndarray): Image to be converted to PIL Image.
        mode (`PIL.Image mode`_): color space and pixel depth of input data (optional).

251
    .. _PIL.Image mode: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#concept-modes
252
253
254
255

    Returns:
        PIL Image: Image converted to PIL Image.
    """
256
257
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(to_pil_image)
258
    if not (isinstance(pic, torch.Tensor) or isinstance(pic, np.ndarray)):
259
        raise TypeError(f"pic should be Tensor or ndarray. Got {type(pic)}.")
260

Varun Agrawal's avatar
Varun Agrawal committed
261
262
    elif isinstance(pic, torch.Tensor):
        if pic.ndimension() not in {2, 3}:
263
            raise ValueError(f"pic should be 2/3 dimensional. Got {pic.ndimension()} dimensions.")
Varun Agrawal's avatar
Varun Agrawal committed
264
265
266

        elif pic.ndimension() == 2:
            # if 2D image, add channel dimension (CHW)
Surgan Jandial's avatar
Surgan Jandial committed
267
            pic = pic.unsqueeze(0)
Varun Agrawal's avatar
Varun Agrawal committed
268

269
270
        # check number of channels
        if pic.shape[-3] > 4:
271
            raise ValueError(f"pic should not have > 4 channels. Got {pic.shape[-3]} channels.")
272

Varun Agrawal's avatar
Varun Agrawal committed
273
274
    elif isinstance(pic, np.ndarray):
        if pic.ndim not in {2, 3}:
275
            raise ValueError(f"pic should be 2/3 dimensional. Got {pic.ndim} dimensions.")
Varun Agrawal's avatar
Varun Agrawal committed
276
277
278
279
280

        elif pic.ndim == 2:
            # if 2D image, add channel dimension (HWC)
            pic = np.expand_dims(pic, 2)

281
282
        # check number of channels
        if pic.shape[-1] > 4:
283
            raise ValueError(f"pic should not have > 4 channels. Got {pic.shape[-1]} channels.")
284

285
    npimg = pic
Varun Agrawal's avatar
Varun Agrawal committed
286
    if isinstance(pic, torch.Tensor):
287
        if pic.is_floating_point() and mode != "F":
288
289
            pic = pic.mul(255).byte()
        npimg = np.transpose(pic.cpu().numpy(), (1, 2, 0))
290
291

    if not isinstance(npimg, np.ndarray):
292
        raise TypeError("Input pic must be a torch.Tensor or NumPy ndarray, not {type(npimg)}")
293
294
295
296
297

    if npimg.shape[2] == 1:
        expected_mode = None
        npimg = npimg[:, :, 0]
        if npimg.dtype == np.uint8:
298
            expected_mode = "L"
vfdev's avatar
vfdev committed
299
        elif npimg.dtype == np.int16:
300
            expected_mode = "I;16"
vfdev's avatar
vfdev committed
301
        elif npimg.dtype == np.int32:
302
            expected_mode = "I"
303
        elif npimg.dtype == np.float32:
304
            expected_mode = "F"
305
        if mode is not None and mode != expected_mode:
306
            raise ValueError(f"Incorrect mode ({mode}) supplied for input type {np.dtype}. Should be {expected_mode}")
307
308
        mode = expected_mode

surgan12's avatar
surgan12 committed
309
    elif npimg.shape[2] == 2:
310
        permitted_2_channel_modes = ["LA"]
surgan12's avatar
surgan12 committed
311
        if mode is not None and mode not in permitted_2_channel_modes:
312
            raise ValueError(f"Only modes {permitted_2_channel_modes} are supported for 2D inputs")
surgan12's avatar
surgan12 committed
313
314

        if mode is None and npimg.dtype == np.uint8:
315
            mode = "LA"
surgan12's avatar
surgan12 committed
316

317
    elif npimg.shape[2] == 4:
318
        permitted_4_channel_modes = ["RGBA", "CMYK", "RGBX"]
319
        if mode is not None and mode not in permitted_4_channel_modes:
320
            raise ValueError(f"Only modes {permitted_4_channel_modes} are supported for 4D inputs")
321
322

        if mode is None and npimg.dtype == np.uint8:
323
            mode = "RGBA"
324
    else:
325
        permitted_3_channel_modes = ["RGB", "YCbCr", "HSV"]
326
        if mode is not None and mode not in permitted_3_channel_modes:
327
            raise ValueError(f"Only modes {permitted_3_channel_modes} are supported for 3D inputs")
328
        if mode is None and npimg.dtype == np.uint8:
329
            mode = "RGB"
330
331

    if mode is None:
332
        raise TypeError(f"Input type {npimg.dtype} is not supported")
333
334
335
336

    return Image.fromarray(npimg, mode=mode)


337
def normalize(tensor: Tensor, mean: List[float], std: List[float], inplace: bool = False) -> Tensor:
338
    """Normalize a float tensor image with mean and standard deviation.
339
    This transform does not support PIL Image.
340

341
    .. note::
surgan12's avatar
surgan12 committed
342
        This transform acts out of place by default, i.e., it does not mutates the input tensor.
343

344
    See :class:`~torchvision.transforms.Normalize` for more details.
345
346

    Args:
347
        tensor (Tensor): Float tensor image of size (C, H, W) or (B, C, H, W) to be normalized.
348
        mean (sequence): Sequence of means for each channel.
349
        std (sequence): Sequence of standard deviations for each channel.
350
        inplace(bool,optional): Bool to make this operation inplace.
351
352
353
354

    Returns:
        Tensor: Normalized Tensor image.
    """
355
356
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(normalize)
357
    if not isinstance(tensor, torch.Tensor):
358
        raise TypeError(f"img should be Tensor Image. Got {type(tensor)}")
359

360
    return F_t.normalize(tensor, mean=mean, std=std, inplace=inplace)
361
362


363
364
365
366
367
368
369
def resize(
    img: Tensor,
    size: List[int],
    interpolation: InterpolationMode = InterpolationMode.BILINEAR,
    max_size: Optional[int] = None,
    antialias: Optional[bool] = None,
) -> Tensor:
vfdev's avatar
vfdev committed
370
    r"""Resize the input image to the given size.
371
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
372
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
373

374
375
376
377
    .. warning::
        The output image might be different depending on its type: when downsampling, the interpolation of PIL images
        and tensors is slightly different, because PIL applies antialiasing. This may lead to significant differences
        in the performance of a network. Therefore, it is preferable to train and serve a model with the same input
378
379
        types. See also below the ``antialias`` parameter, which can help making the output of PIL images and tensors
        closer.
380

381
    Args:
vfdev's avatar
vfdev committed
382
        img (PIL Image or Tensor): Image to be resized.
383
384
        size (sequence or int): Desired output size. If size is a sequence like
            (h, w), the output size will be matched to this. If size is an int,
Vitaliy Chiley's avatar
Vitaliy Chiley committed
385
            the smaller edge of the image will be matched to this number maintaining
386
            the aspect ratio. i.e, if height > width, then image will be rescaled to
vfdev's avatar
vfdev committed
387
            :math:`\left(\text{size} \times \frac{\text{height}}{\text{width}}, \text{size}\right)`.
388
389
390

            .. note::
                In torchscript mode size as single int is not supported, use a sequence of length 1: ``[size, ]``.
391
392
393
394
        interpolation (InterpolationMode): Desired interpolation enum defined by
            :class:`torchvision.transforms.InterpolationMode`.
            Default is ``InterpolationMode.BILINEAR``. If input is Tensor, only ``InterpolationMode.NEAREST``,
            ``InterpolationMode.BILINEAR`` and ``InterpolationMode.BICUBIC`` are supported.
395
396
            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still accepted,
            but deprecated since 0.13 and will be removed in 0.15. Please use InterpolationMode enum.
397
398
399
400
        max_size (int, optional): The maximum allowed for the longer edge of
            the resized image: if the longer edge of the image is greater
            than ``max_size`` after being resized according to ``size``, then
            the image is resized again so that the longer edge is equal to
401
            ``max_size``. As a result, ``size`` might be overruled, i.e the
402
403
404
            smaller edge may be shorter than ``size``. This is only supported
            if ``size`` is an int (or a sequence of length 1 in torchscript
            mode).
405
        antialias (bool, optional): antialias flag. If ``img`` is PIL Image, the flag is ignored and anti-alias
406
            is always used. If ``img`` is Tensor, the flag is False by default and can be set to True for
407
408
            ``InterpolationMode.BILINEAR`` and ``InterpolationMode.BICUBIC`` modes.
            This can help making the output for PIL images and tensors closer.
409
410

    Returns:
vfdev's avatar
vfdev committed
411
        PIL Image or Tensor: Resized image.
412
    """
413
414
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(resize)
415
416
417
    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
418
419
            "Argument 'interpolation' of type int is deprecated since 0.13 and will be removed in 0.15. "
            "Please use InterpolationMode enum."
420
421
422
        )
        interpolation = _interpolation_modes_from_int(interpolation)

423
424
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
425

vfdev's avatar
vfdev committed
426
    if not isinstance(img, torch.Tensor):
427
        if antialias is not None and not antialias:
428
            warnings.warn("Anti-alias option is always applied for PIL Image input. Argument antialias is ignored.")
429
        pil_interpolation = pil_modes_mapping[interpolation]
430
        return F_pil.resize(img, size=size, interpolation=pil_interpolation, max_size=max_size)
vfdev's avatar
vfdev committed
431

432
    return F_t.resize(img, size=size, interpolation=interpolation.value, max_size=max_size, antialias=antialias)
433
434


435
436
def pad(img: Tensor, padding: List[int], fill: int = 0, padding_mode: str = "constant") -> Tensor:
    r"""Pad the given image on all sides with the given "pad" value.
437
    If the image is torch Tensor, it is expected
438
439
440
    to have [..., H, W] shape, where ... means at most 2 leading dimensions for mode reflect and symmetric,
    at most 3 leading dimensions for mode edge,
    and an arbitrary number of leading dimensions for mode constant
441
442

    Args:
443
        img (PIL Image or Tensor): Image to be padded.
444
445
446
        padding (int or sequence): Padding on each border. If a single int is provided this
            is used to pad all borders. If sequence of length 2 is provided this is the padding
            on left/right and top/bottom respectively. If a sequence of length 4 is provided
447
            this is the padding for the left, top, right and bottom borders respectively.
448
449
450
451

            .. note::
                In torchscript mode padding as single int is not supported, use a sequence of
                length 1: ``[padding, ]``.
452
        fill (number or tuple): Pixel fill value for constant fill. Default is 0.
453
454
455
            If a tuple of length 3, it is used to fill R, G, B channels respectively.
            This value is only used when the padding_mode is constant.
            Only number is supported for torch Tensor.
456
            Only int or tuple value is supported for PIL Image.
457
458
        padding_mode (str): Type of padding. Should be: constant, edge, reflect or symmetric.
            Default is constant.
459
460
461

            - constant: pads with a constant value, this value is specified with fill

462
463
            - edge: pads with the last value at the edge of the image.
              If input a 5D torch Tensor, the last 3 dimensions will be padded instead of the last 2
464

465
466
467
            - reflect: pads with reflection of image without repeating the last value on the edge.
              For example, padding [1, 2, 3, 4] with 2 elements on both sides in reflect mode
              will result in [3, 2, 1, 2, 3, 4, 3, 2]
468

469
470
471
            - symmetric: pads with reflection of image repeating the last value on the edge.
              For example, padding [1, 2, 3, 4] with 2 elements on both sides in symmetric mode
              will result in [2, 1, 1, 2, 3, 4, 4, 3]
472
473

    Returns:
474
        PIL Image or Tensor: Padded image.
475
    """
476
477
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(pad)
478
479
    if not isinstance(img, torch.Tensor):
        return F_pil.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
480

481
    return F_t.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
482
483


vfdev's avatar
vfdev committed
484
485
def crop(img: Tensor, top: int, left: int, height: int, width: int) -> Tensor:
    """Crop the given image at specified location and output size.
486
    If the image is torch Tensor, it is expected
487
488
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
    If image size is smaller than output size along any edge, image is padded with 0 and then cropped.
489

490
    Args:
vfdev's avatar
vfdev committed
491
        img (PIL Image or Tensor): Image to be cropped. (0,0) denotes the top left corner of the image.
492
493
494
495
        top (int): Vertical component of the top left corner of the crop box.
        left (int): Horizontal component of the top left corner of the crop box.
        height (int): Height of the crop box.
        width (int): Width of the crop box.
496

497
    Returns:
vfdev's avatar
vfdev committed
498
        PIL Image or Tensor: Cropped image.
499
500
    """

501
502
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(crop)
vfdev's avatar
vfdev committed
503
504
    if not isinstance(img, torch.Tensor):
        return F_pil.crop(img, top, left, height, width)
505

vfdev's avatar
vfdev committed
506
    return F_t.crop(img, top, left, height, width)
507

vfdev's avatar
vfdev committed
508
509
510

def center_crop(img: Tensor, output_size: List[int]) -> Tensor:
    """Crops the given image at the center.
511
    If the image is torch Tensor, it is expected
512
513
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
    If image size is smaller than output size along any edge, image is padded with 0 and then center cropped.
514

515
    Args:
vfdev's avatar
vfdev committed
516
        img (PIL Image or Tensor): Image to be cropped.
517
        output_size (sequence or int): (height, width) of the crop box. If int or sequence with single int,
vfdev's avatar
vfdev committed
518
519
            it is used for both directions.

520
    Returns:
vfdev's avatar
vfdev committed
521
        PIL Image or Tensor: Cropped image.
522
    """
523
524
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(center_crop)
525
526
    if isinstance(output_size, numbers.Number):
        output_size = (int(output_size), int(output_size))
vfdev's avatar
vfdev committed
527
528
529
    elif isinstance(output_size, (tuple, list)) and len(output_size) == 1:
        output_size = (output_size[0], output_size[0])

530
    _, image_height, image_width = get_dimensions(img)
531
    crop_height, crop_width = output_size
vfdev's avatar
vfdev committed
532

533
534
535
536
537
538
539
540
    if crop_width > image_width or crop_height > image_height:
        padding_ltrb = [
            (crop_width - image_width) // 2 if crop_width > image_width else 0,
            (crop_height - image_height) // 2 if crop_height > image_height else 0,
            (crop_width - image_width + 1) // 2 if crop_width > image_width else 0,
            (crop_height - image_height + 1) // 2 if crop_height > image_height else 0,
        ]
        img = pad(img, padding_ltrb, fill=0)  # PIL uses fill value 0
541
        _, image_height, image_width = get_dimensions(img)
542
543
544
        if crop_width == image_width and crop_height == image_height:
            return img

545
546
    crop_top = int(round((image_height - crop_height) / 2.0))
    crop_left = int(round((image_width - crop_width) / 2.0))
547
    return crop(img, crop_top, crop_left, crop_height, crop_width)
548
549


550
def resized_crop(
551
552
553
554
555
556
557
    img: Tensor,
    top: int,
    left: int,
    height: int,
    width: int,
    size: List[int],
    interpolation: InterpolationMode = InterpolationMode.BILINEAR,
558
559
) -> Tensor:
    """Crop the given image and resize it to desired size.
560
    If the image is torch Tensor, it is expected
561
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
562

563
    Notably used in :class:`~torchvision.transforms.RandomResizedCrop`.
564
565

    Args:
566
        img (PIL Image or Tensor): Image to be cropped. (0,0) denotes the top left corner of the image.
567
568
569
570
        top (int): Vertical component of the top left corner of the crop box.
        left (int): Horizontal component of the top left corner of the crop box.
        height (int): Height of the crop box.
        width (int): Width of the crop box.
571
        size (sequence or int): Desired output size. Same semantics as ``resize``.
572
573
574
575
        interpolation (InterpolationMode): Desired interpolation enum defined by
            :class:`torchvision.transforms.InterpolationMode`.
            Default is ``InterpolationMode.BILINEAR``. If input is Tensor, only ``InterpolationMode.NEAREST``,
            ``InterpolationMode.BILINEAR`` and ``InterpolationMode.BICUBIC`` are supported.
576
577
            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still accepted,
            but deprecated since 0.13 and will be removed in 0.15. Please use InterpolationMode enum.
578
    Returns:
579
        PIL Image or Tensor: Cropped image.
580
    """
581
582
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(resized_crop)
583
    img = crop(img, top, left, height, width)
584
585
586
587
    img = resize(img, size, interpolation)
    return img


588
def hflip(img: Tensor) -> Tensor:
589
    """Horizontally flip the given image.
590
591

    Args:
vfdev's avatar
vfdev committed
592
        img (PIL Image or Tensor): Image to be flipped. If img
593
            is a Tensor, it is expected to be in [..., H, W] format,
594
            where ... means it can have an arbitrary number of leading
595
            dimensions.
596
597

    Returns:
vfdev's avatar
vfdev committed
598
        PIL Image or Tensor:  Horizontally flipped image.
599
    """
600
601
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(hflip)
602
603
    if not isinstance(img, torch.Tensor):
        return F_pil.hflip(img)
604

605
    return F_t.hflip(img)
606
607


608
def _get_perspective_coeffs(startpoints: List[List[int]], endpoints: List[List[int]]) -> List[float]:
609
610
    """Helper function to get the coefficients (a, b, c, d, e, f, g, h) for the perspective transforms.

Vitaliy Chiley's avatar
Vitaliy Chiley committed
611
    In Perspective Transform each pixel (x, y) in the original image gets transformed as,
612
613
614
     (x, y) -> ( (ax + by + c) / (gx + hy + 1), (dx + ey + f) / (gx + hy + 1) )

    Args:
615
616
617
618
619
        startpoints (list of list of ints): List containing four lists of two integers corresponding to four corners
            ``[top-left, top-right, bottom-right, bottom-left]`` of the original image.
        endpoints (list of list of ints): List containing four lists of two integers corresponding to four corners
            ``[top-left, top-right, bottom-right, bottom-left]`` of the transformed image.

620
621
622
    Returns:
        octuple (a, b, c, d, e, f, g, h) for transforming each pixel.
    """
623
624
625
626
627
    a_matrix = torch.zeros(2 * len(startpoints), 8, dtype=torch.float)

    for i, (p1, p2) in enumerate(zip(endpoints, startpoints)):
        a_matrix[2 * i, :] = torch.tensor([p1[0], p1[1], 1, 0, 0, 0, -p2[0] * p1[0], -p2[0] * p1[1]])
        a_matrix[2 * i + 1, :] = torch.tensor([0, 0, 0, p1[0], p1[1], 1, -p2[1] * p1[0], -p2[1] * p1[1]])
628

629
    b_matrix = torch.tensor(startpoints, dtype=torch.float).view(8)
630
    res = torch.linalg.lstsq(a_matrix, b_matrix, driver="gels").solution
631

632
    output: List[float] = res.tolist()
633
    return output
634
635


636
def perspective(
637
638
639
640
641
    img: Tensor,
    startpoints: List[List[int]],
    endpoints: List[List[int]],
    interpolation: InterpolationMode = InterpolationMode.BILINEAR,
    fill: Optional[List[float]] = None,
642
643
) -> Tensor:
    """Perform perspective transform of the given image.
644
    If the image is torch Tensor, it is expected
645
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
646
647

    Args:
648
649
650
651
652
        img (PIL Image or Tensor): Image to be transformed.
        startpoints (list of list of ints): List containing four lists of two integers corresponding to four corners
            ``[top-left, top-right, bottom-right, bottom-left]`` of the original image.
        endpoints (list of list of ints): List containing four lists of two integers corresponding to four corners
            ``[top-left, top-right, bottom-right, bottom-left]`` of the transformed image.
653
654
655
        interpolation (InterpolationMode): Desired interpolation enum defined by
            :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.BILINEAR``.
            If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
656
657
            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still accepted,
            but deprecated since 0.13 and will be removed in 0.15. Please use InterpolationMode enum.
658
659
        fill (sequence or number, optional): Pixel fill value for the area outside the transformed
            image. If given a number, the value is used for all bands respectively.
660
661
662
663

            .. note::
                In torchscript mode single int/float value is not supported, please use a sequence
                of length 1: ``[value, ]``.
664

665
    Returns:
666
        PIL Image or Tensor: transformed Image.
667
    """
668
669
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(perspective)
670

671
    coeffs = _get_perspective_coeffs(startpoints, endpoints)
672

673
674
675
    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
676
677
            "Argument 'interpolation' of type int is deprecated since 0.13 and will be removed in 0.15. "
            "Please use InterpolationMode enum."
678
679
680
        )
        interpolation = _interpolation_modes_from_int(interpolation)

681
682
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
683

684
    if not isinstance(img, torch.Tensor):
685
686
        pil_interpolation = pil_modes_mapping[interpolation]
        return F_pil.perspective(img, coeffs, interpolation=pil_interpolation, fill=fill)
687

688
    return F_t.perspective(img, coeffs, interpolation=interpolation.value, fill=fill)
689
690


691
def vflip(img: Tensor) -> Tensor:
692
    """Vertically flip the given image.
693
694

    Args:
vfdev's avatar
vfdev committed
695
        img (PIL Image or Tensor): Image to be flipped. If img
696
            is a Tensor, it is expected to be in [..., H, W] format,
697
            where ... means it can have an arbitrary number of leading
698
            dimensions.
699
700

    Returns:
701
        PIL Image or Tensor:  Vertically flipped image.
702
    """
703
704
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(vflip)
705
706
    if not isinstance(img, torch.Tensor):
        return F_pil.vflip(img)
707

708
    return F_t.vflip(img)
709
710


vfdev's avatar
vfdev committed
711
712
def five_crop(img: Tensor, size: List[int]) -> Tuple[Tensor, Tensor, Tensor, Tensor, Tensor]:
    """Crop the given image into four corners and the central crop.
713
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
714
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
715
716
717
718
719
720

    .. Note::
        This transform returns a tuple of images and there may be a
        mismatch in the number of inputs and targets your ``Dataset`` returns.

    Args:
vfdev's avatar
vfdev committed
721
722
723
        img (PIL Image or Tensor): Image to be cropped.
        size (sequence or int): Desired output size of the crop. If size is an
            int instead of sequence like (h, w), a square crop (size, size) is
724
            made. If provided a sequence of length 1, it will be interpreted as (size[0], size[0]).
725

726
    Returns:
727
       tuple: tuple (tl, tr, bl, br, center)
728
       Corresponding top left, top right, bottom left, bottom right and center crop.
729
    """
730
731
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(five_crop)
732
733
    if isinstance(size, numbers.Number):
        size = (int(size), int(size))
vfdev's avatar
vfdev committed
734
735
    elif isinstance(size, (tuple, list)) and len(size) == 1:
        size = (size[0], size[0])
736

vfdev's avatar
vfdev committed
737
738
739
    if len(size) != 2:
        raise ValueError("Please provide only two dimensions (h, w) for size.")

740
    _, image_height, image_width = get_dimensions(img)
741
742
743
744
745
    crop_height, crop_width = size
    if crop_width > image_width or crop_height > image_height:
        msg = "Requested crop size {} is bigger than input size {}"
        raise ValueError(msg.format(size, (image_height, image_width)))

vfdev's avatar
vfdev committed
746
747
748
749
750
751
752
753
    tl = crop(img, 0, 0, crop_height, crop_width)
    tr = crop(img, 0, image_width - crop_width, crop_height, crop_width)
    bl = crop(img, image_height - crop_height, 0, crop_height, crop_width)
    br = crop(img, image_height - crop_height, image_width - crop_width, crop_height, crop_width)

    center = center_crop(img, [crop_height, crop_width])

    return tl, tr, bl, br, center
754
755


vfdev's avatar
vfdev committed
756
757
758
def ten_crop(img: Tensor, size: List[int], vertical_flip: bool = False) -> List[Tensor]:
    """Generate ten cropped images from the given image.
    Crop the given image into four corners and the central crop plus the
759
    flipped version of these (horizontal flipping is used by default).
760
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
761
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
762
763
764
765
766

    .. Note::
        This transform returns a tuple of images and there may be a
        mismatch in the number of inputs and targets your ``Dataset`` returns.

767
    Args:
vfdev's avatar
vfdev committed
768
        img (PIL Image or Tensor): Image to be cropped.
769
        size (sequence or int): Desired output size of the crop. If size is an
770
            int instead of sequence like (h, w), a square crop (size, size) is
771
            made. If provided a sequence of length 1, it will be interpreted as (size[0], size[0]).
772
        vertical_flip (bool): Use vertical flipping instead of horizontal
773
774

    Returns:
775
        tuple: tuple (tl, tr, bl, br, center, tl_flip, tr_flip, bl_flip, br_flip, center_flip)
776
777
        Corresponding top left, top right, bottom left, bottom right and
        center crop and same for the flipped image.
778
    """
779
780
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(ten_crop)
781
782
    if isinstance(size, numbers.Number):
        size = (int(size), int(size))
vfdev's avatar
vfdev committed
783
784
785
786
787
    elif isinstance(size, (tuple, list)) and len(size) == 1:
        size = (size[0], size[0])

    if len(size) != 2:
        raise ValueError("Please provide only two dimensions (h, w) for size.")
788
789
790
791
792
793
794
795
796
797
798
799

    first_five = five_crop(img, size)

    if vertical_flip:
        img = vflip(img)
    else:
        img = hflip(img)

    second_five = five_crop(img, size)
    return first_five + second_five


800
def adjust_brightness(img: Tensor, brightness_factor: float) -> Tensor:
801
    """Adjust brightness of an image.
802
803

    Args:
vfdev's avatar
vfdev committed
804
        img (PIL Image or Tensor): Image to be adjusted.
805
806
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
            where ... means it can have an arbitrary number of leading dimensions.
807
808
809
810
811
        brightness_factor (float):  How much to adjust the brightness. Can be
            any non negative number. 0 gives a black image, 1 gives the
            original image while 2 increases the brightness by a factor of 2.

    Returns:
vfdev's avatar
vfdev committed
812
        PIL Image or Tensor: Brightness adjusted image.
813
    """
814
815
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(adjust_brightness)
816
817
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_brightness(img, brightness_factor)
818

819
    return F_t.adjust_brightness(img, brightness_factor)
820
821


822
def adjust_contrast(img: Tensor, contrast_factor: float) -> Tensor:
823
    """Adjust contrast of an image.
824
825

    Args:
vfdev's avatar
vfdev committed
826
        img (PIL Image or Tensor): Image to be adjusted.
827
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
828
            where ... means it can have an arbitrary number of leading dimensions.
829
830
831
832
833
        contrast_factor (float): How much to adjust the contrast. Can be any
            non negative number. 0 gives a solid gray image, 1 gives the
            original image while 2 increases the contrast by a factor of 2.

    Returns:
vfdev's avatar
vfdev committed
834
        PIL Image or Tensor: Contrast adjusted image.
835
    """
836
837
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(adjust_contrast)
838
839
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_contrast(img, contrast_factor)
840

841
    return F_t.adjust_contrast(img, contrast_factor)
842
843


844
def adjust_saturation(img: Tensor, saturation_factor: float) -> Tensor:
845
846
847
    """Adjust color saturation of an image.

    Args:
vfdev's avatar
vfdev committed
848
        img (PIL Image or Tensor): Image to be adjusted.
849
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
850
            where ... means it can have an arbitrary number of leading dimensions.
851
852
853
854
855
        saturation_factor (float):  How much to adjust the saturation. 0 will
            give a black and white image, 1 will give the original image while
            2 will enhance the saturation by a factor of 2.

    Returns:
vfdev's avatar
vfdev committed
856
        PIL Image or Tensor: Saturation adjusted image.
857
    """
858
859
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(adjust_saturation)
860
861
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_saturation(img, saturation_factor)
862

863
    return F_t.adjust_saturation(img, saturation_factor)
864
865


866
def adjust_hue(img: Tensor, hue_factor: float) -> Tensor:
867
868
869
870
871
872
873
874
875
    """Adjust hue of an image.

    The image hue is adjusted by converting the image to HSV and
    cyclically shifting the intensities in the hue channel (H).
    The image is then converted back to original image mode.

    `hue_factor` is the amount of shift in H channel and must be in the
    interval `[-0.5, 0.5]`.

876
877
878
    See `Hue`_ for more details.

    .. _Hue: https://en.wikipedia.org/wiki/Hue
879
880

    Args:
881
        img (PIL Image or Tensor): Image to be adjusted.
882
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
883
            where ... means it can have an arbitrary number of leading dimensions.
884
            If img is PIL Image mode "1", "I", "F" and modes with transparency (alpha channel) are not supported.
885
886
887
            Note: the pixel values of the input image has to be non-negative for conversion to HSV space;
            thus it does not work if you normalize your image to an interval with negative values,
            or use an interpolation that generates negative values before using this function.
888
889
890
891
892
893
894
        hue_factor (float):  How much to shift the hue channel. Should be in
            [-0.5, 0.5]. 0.5 and -0.5 give complete reversal of hue channel in
            HSV space in positive and negative direction respectively.
            0 means no shift. Therefore, both -0.5 and 0.5 will give an image
            with complementary colors while 0 gives the original image.

    Returns:
895
        PIL Image or Tensor: Hue adjusted image.
896
    """
897
898
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(adjust_hue)
899
900
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_hue(img, hue_factor)
901

902
    return F_t.adjust_hue(img, hue_factor)
903
904


905
def adjust_gamma(img: Tensor, gamma: float, gain: float = 1) -> Tensor:
906
    r"""Perform gamma correction on an image.
907
908
909
910

    Also known as Power Law Transform. Intensities in RGB mode are adjusted
    based on the following equation:

911
912
913
914
    .. math::
        I_{\text{out}} = 255 \times \text{gain} \times \left(\frac{I_{\text{in}}}{255}\right)^{\gamma}

    See `Gamma Correction`_ for more details.
915

916
    .. _Gamma Correction: https://en.wikipedia.org/wiki/Gamma_correction
917
918

    Args:
919
        img (PIL Image or Tensor): PIL Image to be adjusted.
920
921
922
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
            where ... means it can have an arbitrary number of leading dimensions.
            If img is PIL Image, modes with transparency (alpha channel) are not supported.
923
924
925
        gamma (float): Non negative real number, same as :math:`\gamma` in the equation.
            gamma larger than 1 make the shadows darker,
            while gamma smaller than 1 make dark regions lighter.
926
        gain (float): The constant multiplier.
927
928
    Returns:
        PIL Image or Tensor: Gamma correction adjusted image.
929
    """
930
931
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(adjust_gamma)
932
933
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_gamma(img, gamma, gain)
934

935
    return F_t.adjust_gamma(img, gamma, gain)
936
937


vfdev's avatar
vfdev committed
938
def _get_inverse_affine_matrix(
939
    center: List[float], angle: float, translate: List[float], scale: float, shear: List[float], inverted: bool = True
vfdev's avatar
vfdev committed
940
) -> List[float]:
941
942
    # Helper method to compute inverse matrix for affine transformation

943
944
945
    # Pillow requires inverse affine transformation matrix:
    # Affine matrix is : M = T * C * RotateScaleShear * C^-1
    #
946
947
    # where T is translation matrix: [1, 0, tx | 0, 1, ty | 0, 0, 1]
    #       C is translation matrix to keep center: [1, 0, cx | 0, 1, cy | 0, 0, 1]
948
949
950
    #       RotateScaleShear is rotation with scale and shear matrix
    #
    #       RotateScaleShear(a, s, (sx, sy)) =
951
    #       = R(a) * S(s) * SHy(sy) * SHx(sx)
952
953
    #       = [ s*cos(a - sy)/cos(sy), s*(-cos(a - sy)*tan(sx)/cos(sy) - sin(a)), 0 ]
    #         [ s*sin(a + sy)/cos(sy), s*(-sin(a - sy)*tan(sx)/cos(sy) + cos(a)), 0 ]
954
955
956
957
958
    #         [ 0                    , 0                                      , 1 ]
    # where R is a rotation matrix, S is a scaling matrix, and SHx and SHy are the shears:
    # SHx(s) = [1, -tan(s)] and SHy(s) = [1      , 0]
    #          [0, 1      ]              [-tan(s), 1]
    #
959
    # Thus, the inverse is M^-1 = C * RotateScaleShear^-1 * C^-1 * T^-1
960

961
    rot = math.radians(angle)
962
963
    sx = math.radians(shear[0])
    sy = math.radians(shear[1])
964
965
966
967
968

    cx, cy = center
    tx, ty = translate

    # RSS without scaling
vfdev's avatar
vfdev committed
969
970
971
972
    a = math.cos(rot - sy) / math.cos(sy)
    b = -math.cos(rot - sy) * math.tan(sx) / math.cos(sy) - math.sin(rot)
    c = math.sin(rot - sy) / math.cos(sy)
    d = -math.sin(rot - sy) * math.tan(sx) / math.cos(sy) + math.cos(rot)
973

974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
    if inverted:
        # Inverted rotation matrix with scale and shear
        # det([[a, b], [c, d]]) == 1, since det(rotation) = 1 and det(shear) = 1
        matrix = [d, -b, 0.0, -c, a, 0.0]
        matrix = [x / scale for x in matrix]
        # Apply inverse of translation and of center translation: RSS^-1 * C^-1 * T^-1
        matrix[2] += matrix[0] * (-cx - tx) + matrix[1] * (-cy - ty)
        matrix[5] += matrix[3] * (-cx - tx) + matrix[4] * (-cy - ty)
        # Apply center translation: C * RSS^-1 * C^-1 * T^-1
        matrix[2] += cx
        matrix[5] += cy
    else:
        matrix = [a, b, 0.0, c, d, 0.0]
        matrix = [x * scale for x in matrix]
        # Apply inverse of center translation: RSS * C^-1
        matrix[2] += matrix[0] * (-cx) + matrix[1] * (-cy)
        matrix[5] += matrix[3] * (-cx) + matrix[4] * (-cy)
        # Apply translation and center : T * C * RSS * C^-1
        matrix[2] += cx + tx
        matrix[5] += cy + ty
994

vfdev's avatar
vfdev committed
995
    return matrix
996

vfdev's avatar
vfdev committed
997

vfdev's avatar
vfdev committed
998
def rotate(
999
1000
1001
1002
1003
1004
1005
    img: Tensor,
    angle: float,
    interpolation: InterpolationMode = InterpolationMode.NEAREST,
    expand: bool = False,
    center: Optional[List[int]] = None,
    fill: Optional[List[float]] = None,
    resample: Optional[int] = None,
vfdev's avatar
vfdev committed
1006
1007
) -> Tensor:
    """Rotate the image by angle.
1008
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
1009
1010
1011
1012
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.

    Args:
        img (PIL Image or Tensor): image to be rotated.
1013
        angle (number): rotation angle value in degrees, counter-clockwise.
1014
1015
1016
        interpolation (InterpolationMode): Desired interpolation enum defined by
            :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``.
            If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
1017
1018
            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still accepted,
            but deprecated since 0.13 and will be removed in 0.15. Please use InterpolationMode enum.
vfdev's avatar
vfdev committed
1019
1020
1021
1022
        expand (bool, optional): Optional expansion flag.
            If true, expands the output image to make it large enough to hold the entire rotated image.
            If false or omitted, make the output image the same size as the input image.
            Note that the expand flag assumes rotation around the center and no translation.
1023
        center (sequence, optional): Optional center of rotation. Origin is the upper left corner.
vfdev's avatar
vfdev committed
1024
            Default is the center of the image.
1025
1026
        fill (sequence or number, optional): Pixel fill value for the area outside the transformed
            image. If given a number, the value is used for all bands respectively.
1027
1028
1029
1030

            .. note::
                In torchscript mode single int/float value is not supported, please use a sequence
                of length 1: ``[value, ]``.
1031
1032
1033
1034
        resample (int, optional):
            .. warning::
                This parameter was deprecated in ``0.12`` and will be removed in ``0.14``. Please use ``interpolation``
                instead.
vfdev's avatar
vfdev committed
1035
1036
1037
1038
1039
1040
1041

    Returns:
        PIL Image or Tensor: Rotated image.

    .. _filters: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#filters

    """
1042
1043
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(rotate)
1044
1045
    if resample is not None:
        warnings.warn(
1046
1047
            "The parameter 'resample' is deprecated since 0.12 and will be removed 0.14. "
            "Please use 'interpolation' instead."
1048
1049
1050
1051
1052
1053
        )
        interpolation = _interpolation_modes_from_int(resample)

    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
1054
1055
            "Argument 'interpolation' of type int is deprecated since 0.13 and will be removed in 0.15. "
            "Please use InterpolationMode enum."
1056
1057
1058
        )
        interpolation = _interpolation_modes_from_int(interpolation)

vfdev's avatar
vfdev committed
1059
1060
1061
1062
1063
1064
    if not isinstance(angle, (int, float)):
        raise TypeError("Argument angle should be int or float")

    if center is not None and not isinstance(center, (list, tuple)):
        raise TypeError("Argument center should be a sequence")

1065
1066
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
1067

vfdev's avatar
vfdev committed
1068
    if not isinstance(img, torch.Tensor):
1069
1070
        pil_interpolation = pil_modes_mapping[interpolation]
        return F_pil.rotate(img, angle=angle, interpolation=pil_interpolation, expand=expand, center=center, fill=fill)
vfdev's avatar
vfdev committed
1071
1072
1073

    center_f = [0.0, 0.0]
    if center is not None:
1074
        _, height, width = get_dimensions(img)
1075
        # Center values should be in pixel coordinates but translated such that (0, 0) corresponds to image center.
1076
        center_f = [1.0 * (c - s * 0.5) for c, s in zip(center, [width, height])]
1077

vfdev's avatar
vfdev committed
1078
1079
1080
    # due to current incoherence of rotation angle direction between affine and rotate implementations
    # we need to set -angle.
    matrix = _get_inverse_affine_matrix(center_f, -angle, [0.0, 0.0], 1.0, [0.0, 0.0])
1081
    return F_t.rotate(img, matrix=matrix, interpolation=interpolation.value, expand=expand, fill=fill)
vfdev's avatar
vfdev committed
1082
1083


vfdev's avatar
vfdev committed
1084
def affine(
1085
1086
1087
1088
1089
1090
1091
1092
1093
    img: Tensor,
    angle: float,
    translate: List[int],
    scale: float,
    shear: List[float],
    interpolation: InterpolationMode = InterpolationMode.NEAREST,
    fill: Optional[List[float]] = None,
    resample: Optional[int] = None,
    fillcolor: Optional[List[float]] = None,
1094
    center: Optional[List[int]] = None,
vfdev's avatar
vfdev committed
1095
1096
) -> Tensor:
    """Apply affine transformation on the image keeping image center invariant.
1097
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
1098
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
1099
1100

    Args:
vfdev's avatar
vfdev committed
1101
        img (PIL Image or Tensor): image to transform.
1102
1103
        angle (number): rotation angle in degrees between -180 and 180, clockwise direction.
        translate (sequence of integers): horizontal and vertical translations (post-rotation translation)
1104
        scale (float): overall scale
1105
1106
        shear (float or sequence): shear angle value in degrees between -180 to 180, clockwise direction.
            If a sequence is specified, the first value corresponds to a shear parallel to the x axis, while
vfdev's avatar
vfdev committed
1107
            the second value corresponds to a shear parallel to the y axis.
1108
1109
1110
        interpolation (InterpolationMode): Desired interpolation enum defined by
            :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``.
            If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.
1111
1112
            For backward compatibility integer values (e.g. ``PIL.Image[.Resampling].NEAREST``) are still accepted,
            but deprecated since 0.13 and will be removed in 0.15. Please use InterpolationMode enum.
1113
1114
        fill (sequence or number, optional): Pixel fill value for the area outside the transformed
            image. If given a number, the value is used for all bands respectively.
1115
1116
1117
1118

            .. note::
                In torchscript mode single int/float value is not supported, please use a sequence
                of length 1: ``[value, ]``.
1119
1120
1121
1122
1123
1124
1125
        fillcolor (sequence or number, optional):
            .. warning::
                This parameter was deprecated in ``0.12`` and will be removed in ``0.14``. Please use ``fill`` instead.
        resample (int, optional):
            .. warning::
                This parameter was deprecated in ``0.12`` and will be removed in ``0.14``. Please use ``interpolation``
                instead.
1126
1127
        center (sequence, optional): Optional center of rotation. Origin is the upper left corner.
            Default is the center of the image.
vfdev's avatar
vfdev committed
1128
1129
1130

    Returns:
        PIL Image or Tensor: Transformed image.
1131
    """
1132
1133
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(affine)
1134
1135
    if resample is not None:
        warnings.warn(
1136
1137
            "The parameter 'resample' is deprecated since 0.12 and will be removed in 0.14. "
            "Please use 'interpolation' instead."
1138
1139
1140
1141
1142
1143
        )
        interpolation = _interpolation_modes_from_int(resample)

    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
1144
1145
            "Argument 'interpolation' of type int is deprecated since 0.13 and will be removed in 0.15. "
            "Please use InterpolationMode enum."
1146
1147
1148
1149
        )
        interpolation = _interpolation_modes_from_int(interpolation)

    if fillcolor is not None:
1150
1151
1152
1153
        warnings.warn(
            "The parameter 'fillcolor' is deprecated since 0.12 and will be removed in 0.14. "
            "Please use 'fill' instead."
        )
1154
1155
        fill = fillcolor

vfdev's avatar
vfdev committed
1156
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
    if not isinstance(angle, (int, float)):
        raise TypeError("Argument angle should be int or float")

    if not isinstance(translate, (list, tuple)):
        raise TypeError("Argument translate should be a sequence")

    if len(translate) != 2:
        raise ValueError("Argument translate should be a sequence of length 2")

    if scale <= 0.0:
        raise ValueError("Argument scale should be positive")

    if not isinstance(shear, (numbers.Number, (list, tuple))):
        raise TypeError("Shear should be either a single value or a sequence of two values")

1171
1172
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
1173

vfdev's avatar
vfdev committed
1174
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
    if isinstance(angle, int):
        angle = float(angle)

    if isinstance(translate, tuple):
        translate = list(translate)

    if isinstance(shear, numbers.Number):
        shear = [shear, 0.0]

    if isinstance(shear, tuple):
        shear = list(shear)

    if len(shear) == 1:
        shear = [shear[0], shear[0]]

    if len(shear) != 2:
1190
        raise ValueError(f"Shear should be a sequence containing two values. Got {shear}")
vfdev's avatar
vfdev committed
1191

1192
1193
1194
    if center is not None and not isinstance(center, (list, tuple)):
        raise TypeError("Argument center should be a sequence")

1195
    _, height, width = get_dimensions(img)
vfdev's avatar
vfdev committed
1196
    if not isinstance(img, torch.Tensor):
1197
        # center = (width * 0.5 + 0.5, height * 0.5 + 0.5)
vfdev's avatar
vfdev committed
1198
1199
        # it is visually better to estimate the center without 0.5 offset
        # otherwise image rotated by 90 degrees is shifted vs output image of torch.rot90 or F_t.affine
1200
        if center is None:
1201
            center = [width * 0.5, height * 0.5]
vfdev's avatar
vfdev committed
1202
        matrix = _get_inverse_affine_matrix(center, angle, translate, scale, shear)
1203
1204
        pil_interpolation = pil_modes_mapping[interpolation]
        return F_pil.affine(img, matrix=matrix, interpolation=pil_interpolation, fill=fill)
1205

1206
1207
    center_f = [0.0, 0.0]
    if center is not None:
1208
        _, height, width = get_dimensions(img)
1209
        # Center values should be in pixel coordinates but translated such that (0, 0) corresponds to image center.
1210
        center_f = [1.0 * (c - s * 0.5) for c, s in zip(center, [width, height])]
1211

1212
    translate_f = [1.0 * t for t in translate]
1213
    matrix = _get_inverse_affine_matrix(center_f, angle, translate_f, scale, shear)
1214
    return F_t.affine(img, matrix=matrix, interpolation=interpolation.value, fill=fill)
1215
1216


1217
@torch.jit.unused
1218
def to_grayscale(img, num_output_channels=1):
1219
    """Convert PIL image of any mode (RGB, HSV, LAB, etc) to grayscale version of image.
1220
    This transform does not support torch Tensor.
1221
1222

    Args:
1223
        img (PIL Image): PIL Image to be converted to grayscale.
1224
        num_output_channels (int): number of channels of the output image. Value can be 1 or 3. Default is 1.
1225
1226

    Returns:
1227
1228
        PIL Image: Grayscale version of the image.

1229
1230
        - if num_output_channels = 1 : returned image is single channel
        - if num_output_channels = 3 : returned image is 3 channel with r = g = b
1231
    """
1232
1233
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(to_grayscale)
1234
1235
    if isinstance(img, Image.Image):
        return F_pil.to_grayscale(img, num_output_channels)
1236

1237
1238
1239
1240
1241
    raise TypeError("Input should be PIL Image")


def rgb_to_grayscale(img: Tensor, num_output_channels: int = 1) -> Tensor:
    """Convert RGB image to grayscale version of image.
1242
1243
    If the image is torch Tensor, it is expected
    to have [..., 3, H, W] shape, where ... means an arbitrary number of leading dimensions
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255

    Note:
        Please, note that this method supports only RGB images as input. For inputs in other color spaces,
        please, consider using meth:`~torchvision.transforms.functional.to_grayscale` with PIL Image.

    Args:
        img (PIL Image or Tensor): RGB Image to be converted to grayscale.
        num_output_channels (int): number of channels of the output image. Value can be 1 or 3. Default, 1.

    Returns:
        PIL Image or Tensor: Grayscale version of the image.

1256
1257
        - if num_output_channels = 1 : returned image is single channel
        - if num_output_channels = 3 : returned image is 3 channel with r = g = b
1258
    """
1259
1260
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(rgb_to_grayscale)
1261
1262
1263
1264
    if not isinstance(img, torch.Tensor):
        return F_pil.to_grayscale(img, num_output_channels)

    return F_t.rgb_to_grayscale(img, num_output_channels)
1265
1266


1267
def erase(img: Tensor, i: int, j: int, h: int, w: int, v: Tensor, inplace: bool = False) -> Tensor:
1268
    """Erase the input Tensor Image with given value.
1269
    This transform does not support PIL Image.
1270
1271
1272
1273
1274
1275
1276
1277

    Args:
        img (Tensor Image): Tensor image of size (C, H, W) to be erased
        i (int): i in (i,j) i.e coordinates of the upper left corner.
        j (int): j in (i,j) i.e coordinates of the upper left corner.
        h (int): Height of the erased region.
        w (int): Width of the erased region.
        v: Erasing value.
Zhun Zhong's avatar
Zhun Zhong committed
1278
        inplace(bool, optional): For in-place operations. By default is set False.
1279
1280
1281
1282

    Returns:
        Tensor Image: Erased image.
    """
1283
1284
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(erase)
1285
    if not isinstance(img, torch.Tensor):
1286
        raise TypeError(f"img should be Tensor Image. Got {type(img)}")
1287

1288
    return F_t.erase(img, i, j, h, w, v, inplace=inplace)
1289
1290
1291


def gaussian_blur(img: Tensor, kernel_size: List[int], sigma: Optional[List[float]] = None) -> Tensor:
1292
1293
1294
    """Performs Gaussian blurring on the image by given kernel.
    If the image is torch Tensor, it is expected
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
1295
1296
1297
1298
1299

    Args:
        img (PIL Image or Tensor): Image to be blurred
        kernel_size (sequence of ints or int): Gaussian kernel size. Can be a sequence of integers
            like ``(kx, ky)`` or a single integer for square kernels.
1300
1301
1302
1303

            .. note::
                In torchscript mode kernel_size as single int is not supported, use a sequence of
                length 1: ``[ksize, ]``.
1304
1305
1306
1307
        sigma (sequence of floats or float, optional): Gaussian kernel standard deviation. Can be a
            sequence of floats like ``(sigma_x, sigma_y)`` or a single float to define the
            same sigma in both X/Y directions. If None, then it is computed using
            ``kernel_size`` as ``sigma = 0.3 * ((kernel_size - 1) * 0.5 - 1) + 0.8``.
1308
1309
1310
1311
1312
            Default, None.

            .. note::
                In torchscript mode sigma as single float is
                not supported, use a sequence of length 1: ``[sigma, ]``.
1313
1314
1315
1316

    Returns:
        PIL Image or Tensor: Gaussian Blurred version of the image.
    """
1317
1318
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(gaussian_blur)
1319
    if not isinstance(kernel_size, (int, list, tuple)):
1320
        raise TypeError(f"kernel_size should be int or a sequence of integers. Got {type(kernel_size)}")
1321
1322
1323
    if isinstance(kernel_size, int):
        kernel_size = [kernel_size, kernel_size]
    if len(kernel_size) != 2:
1324
        raise ValueError(f"If kernel_size is a sequence its length should be 2. Got {len(kernel_size)}")
1325
1326
    for ksize in kernel_size:
        if ksize % 2 == 0 or ksize < 0:
1327
            raise ValueError(f"kernel_size should have odd and positive integers. Got {kernel_size}")
1328
1329
1330
1331
1332

    if sigma is None:
        sigma = [ksize * 0.15 + 0.35 for ksize in kernel_size]

    if sigma is not None and not isinstance(sigma, (int, float, list, tuple)):
1333
        raise TypeError(f"sigma should be either float or sequence of floats. Got {type(sigma)}")
1334
1335
1336
1337
1338
    if isinstance(sigma, (int, float)):
        sigma = [float(sigma), float(sigma)]
    if isinstance(sigma, (list, tuple)) and len(sigma) == 1:
        sigma = [sigma[0], sigma[0]]
    if len(sigma) != 2:
1339
        raise ValueError(f"If sigma is a sequence, its length should be 2. Got {len(sigma)}")
1340
    for s in sigma:
1341
        if s <= 0.0:
1342
            raise ValueError(f"sigma should have positive values. Got {sigma}")
1343
1344
1345
1346

    t_img = img
    if not isinstance(img, torch.Tensor):
        if not F_pil._is_pil_image(img):
1347
            raise TypeError(f"img should be PIL Image or Tensor. Got {type(img)}")
1348

1349
        t_img = pil_to_tensor(img)
1350
1351
1352
1353

    output = F_t.gaussian_blur(t_img, kernel_size, sigma)

    if not isinstance(img, torch.Tensor):
1354
        output = to_pil_image(output, mode=img.mode)
1355
    return output
1356
1357
1358


def invert(img: Tensor) -> Tensor:
1359
    """Invert the colors of an RGB/grayscale image.
1360
1361
1362

    Args:
        img (PIL Image or Tensor): Image to have its colors inverted.
1363
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1364
1365
            where ... means it can have an arbitrary number of leading dimensions.
            If img is PIL Image, it is expected to be in mode "L" or "RGB".
1366
1367
1368
1369

    Returns:
        PIL Image or Tensor: Color inverted image.
    """
1370
1371
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(invert)
1372
1373
1374
1375
1376
1377
1378
    if not isinstance(img, torch.Tensor):
        return F_pil.invert(img)

    return F_t.invert(img)


def posterize(img: Tensor, bits: int) -> Tensor:
1379
    """Posterize an image by reducing the number of bits for each color channel.
1380
1381
1382

    Args:
        img (PIL Image or Tensor): Image to have its colors posterized.
1383
            If img is torch Tensor, it should be of type torch.uint8 and
1384
1385
1386
            it is expected to be in [..., 1 or 3, H, W] format, where ... means
            it can have an arbitrary number of leading dimensions.
            If img is PIL Image, it is expected to be in mode "L" or "RGB".
1387
1388
1389
1390
        bits (int): The number of bits to keep for each channel (0-8).
    Returns:
        PIL Image or Tensor: Posterized image.
    """
1391
1392
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(posterize)
1393
    if not (0 <= bits <= 8):
1394
        raise ValueError(f"The number if bits should be between 0 and 8. Got {bits}")
1395
1396
1397
1398
1399
1400
1401
1402

    if not isinstance(img, torch.Tensor):
        return F_pil.posterize(img, bits)

    return F_t.posterize(img, bits)


def solarize(img: Tensor, threshold: float) -> Tensor:
1403
    """Solarize an RGB/grayscale image by inverting all pixel values above a threshold.
1404
1405
1406

    Args:
        img (PIL Image or Tensor): Image to have its colors inverted.
1407
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1408
1409
            where ... means it can have an arbitrary number of leading dimensions.
            If img is PIL Image, it is expected to be in mode "L" or "RGB".
1410
1411
1412
1413
        threshold (float): All pixels equal or above this value are inverted.
    Returns:
        PIL Image or Tensor: Solarized image.
    """
1414
1415
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(solarize)
1416
1417
1418
1419
1420
1421
1422
    if not isinstance(img, torch.Tensor):
        return F_pil.solarize(img, threshold)

    return F_t.solarize(img, threshold)


def adjust_sharpness(img: Tensor, sharpness_factor: float) -> Tensor:
1423
    """Adjust the sharpness of an image.
1424
1425
1426

    Args:
        img (PIL Image or Tensor): Image to be adjusted.
1427
1428
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
            where ... means it can have an arbitrary number of leading dimensions.
1429
1430
1431
1432
1433
1434
1435
        sharpness_factor (float):  How much to adjust the sharpness. Can be
            any non negative number. 0 gives a blurred image, 1 gives the
            original image while 2 increases the sharpness by a factor of 2.

    Returns:
        PIL Image or Tensor: Sharpness adjusted image.
    """
1436
1437
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(adjust_sharpness)
1438
1439
1440
1441
1442
1443
1444
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_sharpness(img, sharpness_factor)

    return F_t.adjust_sharpness(img, sharpness_factor)


def autocontrast(img: Tensor) -> Tensor:
1445
    """Maximize contrast of an image by remapping its
1446
1447
1448
1449
1450
    pixels per channel so that the lowest becomes black and the lightest
    becomes white.

    Args:
        img (PIL Image or Tensor): Image on which autocontrast is applied.
1451
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1452
1453
            where ... means it can have an arbitrary number of leading dimensions.
            If img is PIL Image, it is expected to be in mode "L" or "RGB".
1454
1455
1456
1457

    Returns:
        PIL Image or Tensor: An image that was autocontrasted.
    """
1458
1459
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(autocontrast)
1460
1461
1462
1463
1464
1465
1466
    if not isinstance(img, torch.Tensor):
        return F_pil.autocontrast(img)

    return F_t.autocontrast(img)


def equalize(img: Tensor) -> Tensor:
1467
    """Equalize the histogram of an image by applying
1468
1469
1470
1471
1472
    a non-linear mapping to the input in order to create a uniform
    distribution of grayscale values in the output.

    Args:
        img (PIL Image or Tensor): Image on which equalize is applied.
1473
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1474
            where ... means it can have an arbitrary number of leading dimensions.
1475
            The tensor dtype must be ``torch.uint8`` and values are expected to be in ``[0, 255]``.
1476
            If img is PIL Image, it is expected to be in mode "P", "L" or "RGB".
1477
1478
1479
1480

    Returns:
        PIL Image or Tensor: An image that was equalized.
    """
1481
1482
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(equalize)
1483
1484
1485
1486
    if not isinstance(img, torch.Tensor):
        return F_pil.equalize(img)

    return F_t.equalize(img)
1487
1488
1489
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508
1509
1510
1511
1512
1513
1514
1515
1516
1517
1518
1519
1520
1521
1522
1523
1524
1525
1526
1527
1528
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547
1548
1549
1550


def elastic_transform(
    img: Tensor,
    displacement: Tensor,
    interpolation: InterpolationMode = InterpolationMode.BILINEAR,
    fill: Optional[List[float]] = None,
) -> Tensor:
    """Transform a tensor image with elastic transformations.
    Given alpha and sigma, it will generate displacement
    vectors for all pixels based on random offsets. Alpha controls the strength
    and sigma controls the smoothness of the displacements.
    The displacements are added to an identity grid and the resulting grid is
    used to grid_sample from the image.

    Applications:
        Randomly transforms the morphology of objects in images and produces a
        see-through-water-like effect.

    Args:
        img (PIL Image or Tensor): Image on which elastic_transform is applied.
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
            where ... means it can have an arbitrary number of leading dimensions.
            If img is PIL Image, it is expected to be in mode "P", "L" or "RGB".
        displacement (Tensor): The displacement field.
        interpolation (InterpolationMode): Desired interpolation enum defined by
            :class:`torchvision.transforms.InterpolationMode`.
            Default is ``InterpolationMode.BILINEAR``.
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
        fill (number or str or tuple): Pixel fill value for constant fill. Default is 0.
            If a tuple of length 3, it is used to fill R, G, B channels respectively.
            This value is only used when the padding_mode is constant.
            Only number is supported for torch Tensor.
            Only int or str or tuple value is supported for PIL Image.
    """
    if not torch.jit.is_scripting() and not torch.jit.is_tracing():
        _log_api_usage_once(elastic_transform)
    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
        )
        interpolation = _interpolation_modes_from_int(interpolation)

    if not isinstance(displacement, torch.Tensor):
        raise TypeError("displacement should be a Tensor")

    t_img = img
    if not isinstance(img, torch.Tensor):
        if not F_pil._is_pil_image(img):
            raise TypeError(f"img should be PIL Image or Tensor. Got {type(img)}")
        t_img = pil_to_tensor(img)

    output = F_t.elastic_transform(
        t_img,
        displacement,
        interpolation=interpolation.value,
        fill=fill,
    )

    if not isinstance(img, torch.Tensor):
        output = to_pil_image(output, mode=img.mode)
    return output