functional.py 56.1 KB
Newer Older
1
import math
2
3
import numbers
import warnings
4
from enum import Enum
5
6

import numpy as np
vfdev's avatar
vfdev committed
7
from PIL import Image
8
9
10

import torch
from torch import Tensor
11
from typing import List, Tuple, Any, Optional
12

13
14
15
16
17
try:
    import accimage
except ImportError:
    accimage = None

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
    """
    NEAREST = "nearest"
    BILINEAR = "bilinear"
    BICUBIC = "bicubic"
    # For PIL compatibility
    BOX = "box"
    HAMMING = "hamming"
    LANCZOS = "lanczos"


# TODO: Once torchscript supports Enums with staticmethod
36
37
# this can be put into InterpolationMode as staticmethod
def _interpolation_modes_from_int(i: int) -> InterpolationMode:
38
    inverse_modes_mapping = {
39
40
41
42
43
44
        0: InterpolationMode.NEAREST,
        2: InterpolationMode.BILINEAR,
        3: InterpolationMode.BICUBIC,
        4: InterpolationMode.BOX,
        5: InterpolationMode.HAMMING,
        1: InterpolationMode.LANCZOS,
45
46
47
48
49
    }
    return inverse_modes_mapping[i]


pil_modes_mapping = {
50
51
52
53
54
55
    InterpolationMode.NEAREST: 0,
    InterpolationMode.BILINEAR: 2,
    InterpolationMode.BICUBIC: 3,
    InterpolationMode.BOX: 4,
    InterpolationMode.HAMMING: 5,
    InterpolationMode.LANCZOS: 1,
56
57
}

vfdev's avatar
vfdev committed
58
59
60
61
_is_pil_image = F_pil._is_pil_image


def _get_image_size(img: Tensor) -> List[int]:
62
    """Returns image size as [w, h]
vfdev's avatar
vfdev committed
63
64
65
    """
    if isinstance(img, torch.Tensor):
        return F_t._get_image_size(img)
66

vfdev's avatar
vfdev committed
67
    return F_pil._get_image_size(img)
68

vfdev's avatar
vfdev committed
69

70
def _get_image_num_channels(img: Tensor) -> int:
71
72
    """Returns number of image channels
    """
73
74
75
76
77
78
    if isinstance(img, torch.Tensor):
        return F_t._get_image_num_channels(img)

    return F_pil._get_image_num_channels(img)


vfdev's avatar
vfdev committed
79
80
@torch.jit.unused
def _is_numpy(img: Any) -> bool:
81
82
83
    return isinstance(img, np.ndarray)


vfdev's avatar
vfdev committed
84
85
@torch.jit.unused
def _is_numpy_image(img: Any) -> bool:
86
    return img.ndim in {2, 3}
87
88
89
90


def to_tensor(pic):
    """Convert a ``PIL Image`` or ``numpy.ndarray`` to tensor.
91
    This function does not support torchscript.
92

93
    See :class:`~torchvision.transforms.ToTensor` for more details.
94
95
96
97
98
99
100

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

    Returns:
        Tensor: Converted image.
    """
vfdev's avatar
vfdev committed
101
    if not(F_pil._is_pil_image(pic) or _is_numpy(pic)):
102
103
        raise TypeError('pic should be PIL Image or ndarray. Got {}'.format(type(pic)))

104
105
106
    if _is_numpy(pic) and not _is_numpy_image(pic):
        raise ValueError('pic should be 2/3 dimensional. Got {} dimensions.'.format(pic.ndim))

107
108
    default_float_dtype = torch.get_default_dtype()

109
110
    if isinstance(pic, np.ndarray):
        # handle numpy array
surgan12's avatar
surgan12 committed
111
112
113
        if pic.ndim == 2:
            pic = pic[:, :, None]

114
        img = torch.from_numpy(pic.transpose((2, 0, 1))).contiguous()
115
        # backward compatibility
116
        if isinstance(img, torch.ByteTensor):
117
            return img.to(dtype=default_float_dtype).div(255)
118
119
        else:
            return img
120
121

    if accimage is not None and isinstance(pic, accimage.Image):
122
        nppic = np.zeros([pic.channels, pic.height, pic.width], dtype=np.float32)
123
        pic.copyto(nppic)
124
        return torch.from_numpy(nppic).to(dtype=default_float_dtype)
125
126

    # handle PIL Image
127
128
129
130
    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)
    )
131

132
133
    if pic.mode == '1':
        img = 255 * img
134
    img = img.view(pic.size[1], pic.size[0], len(pic.getbands()))
135
    # put it from HWC to CHW format
136
    img = img.permute((2, 0, 1)).contiguous()
137
    if isinstance(img, torch.ByteTensor):
138
        return img.to(dtype=default_float_dtype).div(255)
139
140
141
142
    else:
        return img


143
144
def pil_to_tensor(pic):
    """Convert a ``PIL Image`` to a tensor of the same type.
145
    This function does not support torchscript.
146

vfdev's avatar
vfdev committed
147
    See :class:`~torchvision.transforms.PILToTensor` for more details.
148
149
150
151
152
153
154

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

    Returns:
        Tensor: Converted image.
    """
155
    if not F_pil._is_pil_image(pic):
156
157
158
        raise TypeError('pic should be PIL Image. Got {}'.format(type(pic)))

    if accimage is not None and isinstance(pic, accimage.Image):
159
160
        # accimage format is always uint8 internally, so always return uint8 here
        nppic = np.zeros([pic.channels, pic.height, pic.width], dtype=np.uint8)
161
162
163
164
165
166
167
168
169
170
171
        pic.copyto(nppic)
        return torch.as_tensor(nppic)

    # handle PIL Image
    img = torch.as_tensor(np.asarray(pic))
    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


172
173
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
174
    This function does not support PIL Image.
175
176
177
178
179
180

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

    Returns:
vfdev's avatar
vfdev committed
181
        Tensor: Converted image
182
183
184
185
186
187
188
189
190
191
192
193

    .. 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``.
    """
194
195
196
197
    if not isinstance(image, torch.Tensor):
        raise TypeError('Input img should be Tensor Image')

    return F_t.convert_image_dtype(image, dtype)
198
199


200
def to_pil_image(pic, mode=None):
201
    """Convert a tensor or an ndarray to PIL Image. This function does not support torchscript.
202

203
    See :class:`~torchvision.transforms.ToPILImage` for more details.
204
205
206
207
208

    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).

209
    .. _PIL.Image mode: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#concept-modes
210
211
212
213

    Returns:
        PIL Image: Image converted to PIL Image.
    """
Varun Agrawal's avatar
Varun Agrawal committed
214
    if not(isinstance(pic, torch.Tensor) or isinstance(pic, np.ndarray)):
215
216
        raise TypeError('pic should be Tensor or ndarray. Got {}.'.format(type(pic)))

Varun Agrawal's avatar
Varun Agrawal committed
217
218
219
220
221
222
    elif isinstance(pic, torch.Tensor):
        if pic.ndimension() not in {2, 3}:
            raise ValueError('pic should be 2/3 dimensional. Got {} dimensions.'.format(pic.ndimension()))

        elif pic.ndimension() == 2:
            # if 2D image, add channel dimension (CHW)
Surgan Jandial's avatar
Surgan Jandial committed
223
            pic = pic.unsqueeze(0)
Varun Agrawal's avatar
Varun Agrawal committed
224

225
226
227
228
        # check number of channels
        if pic.shape[-3] > 4:
            raise ValueError('pic should not have > 4 channels. Got {} channels.'.format(pic.shape[-3]))

Varun Agrawal's avatar
Varun Agrawal committed
229
230
231
232
233
234
235
236
    elif isinstance(pic, np.ndarray):
        if pic.ndim not in {2, 3}:
            raise ValueError('pic should be 2/3 dimensional. Got {} dimensions.'.format(pic.ndim))

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

237
238
239
240
        # check number of channels
        if pic.shape[-1] > 4:
            raise ValueError('pic should not have > 4 channels. Got {} channels.'.format(pic.shape[-1]))

241
    npimg = pic
Varun Agrawal's avatar
Varun Agrawal committed
242
    if isinstance(pic, torch.Tensor):
243
244
245
        if pic.is_floating_point() and mode != 'F':
            pic = pic.mul(255).byte()
        npimg = np.transpose(pic.cpu().numpy(), (1, 2, 0))
246
247
248
249
250
251
252
253
254
255

    if not isinstance(npimg, np.ndarray):
        raise TypeError('Input pic must be a torch.Tensor or NumPy ndarray, ' +
                        'not {}'.format(type(npimg)))

    if npimg.shape[2] == 1:
        expected_mode = None
        npimg = npimg[:, :, 0]
        if npimg.dtype == np.uint8:
            expected_mode = 'L'
vfdev's avatar
vfdev committed
256
        elif npimg.dtype == np.int16:
257
            expected_mode = 'I;16'
vfdev's avatar
vfdev committed
258
        elif npimg.dtype == np.int32:
259
260
261
262
263
264
265
266
            expected_mode = 'I'
        elif npimg.dtype == np.float32:
            expected_mode = 'F'
        if mode is not None and mode != expected_mode:
            raise ValueError("Incorrect mode ({}) supplied for input type {}. Should be {}"
                             .format(mode, np.dtype, expected_mode))
        mode = expected_mode

surgan12's avatar
surgan12 committed
267
268
269
270
271
272
273
274
    elif npimg.shape[2] == 2:
        permitted_2_channel_modes = ['LA']
        if mode is not None and mode not in permitted_2_channel_modes:
            raise ValueError("Only modes {} are supported for 2D inputs".format(permitted_2_channel_modes))

        if mode is None and npimg.dtype == np.uint8:
            mode = 'LA'

275
    elif npimg.shape[2] == 4:
surgan12's avatar
surgan12 committed
276
        permitted_4_channel_modes = ['RGBA', 'CMYK', 'RGBX']
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
        if mode is not None and mode not in permitted_4_channel_modes:
            raise ValueError("Only modes {} are supported for 4D inputs".format(permitted_4_channel_modes))

        if mode is None and npimg.dtype == np.uint8:
            mode = 'RGBA'
    else:
        permitted_3_channel_modes = ['RGB', 'YCbCr', 'HSV']
        if mode is not None and mode not in permitted_3_channel_modes:
            raise ValueError("Only modes {} are supported for 3D inputs".format(permitted_3_channel_modes))
        if mode is None and npimg.dtype == np.uint8:
            mode = 'RGB'

    if mode is None:
        raise TypeError('Input type {} is not supported'.format(npimg.dtype))

    return Image.fromarray(npimg, mode=mode)


295
def normalize(tensor: Tensor, mean: List[float], std: List[float], inplace: bool = False) -> Tensor:
296
    """Normalize a float tensor image with mean and standard deviation.
297
    This transform does not support PIL Image.
298

299
    .. note::
surgan12's avatar
surgan12 committed
300
        This transform acts out of place by default, i.e., it does not mutates the input tensor.
301

302
    See :class:`~torchvision.transforms.Normalize` for more details.
303
304

    Args:
305
        tensor (Tensor): Float tensor image of size (C, H, W) or (B, C, H, W) to be normalized.
306
        mean (sequence): Sequence of means for each channel.
307
        std (sequence): Sequence of standard deviations for each channel.
308
        inplace(bool,optional): Bool to make this operation inplace.
309
310
311
312

    Returns:
        Tensor: Normalized Tensor image.
    """
313
314
    if not isinstance(tensor, torch.Tensor):
        raise TypeError('Input tensor should be a torch tensor. Got {}.'.format(type(tensor)))
315

316
317
318
    if not tensor.is_floating_point():
        raise TypeError('Input tensor should be a float tensor. Got {}.'.format(tensor.dtype))

319
320
    if tensor.ndim < 3:
        raise ValueError('Expected tensor to be a tensor image of size (..., C, H, W). Got tensor.size() = '
321
                         '{}.'.format(tensor.size()))
322

surgan12's avatar
surgan12 committed
323
324
325
    if not inplace:
        tensor = tensor.clone()

326
327
328
    dtype = tensor.dtype
    mean = torch.as_tensor(mean, dtype=dtype, device=tensor.device)
    std = torch.as_tensor(std, dtype=dtype, device=tensor.device)
329
330
    if (std == 0).any():
        raise ValueError('std evaluated to zero after conversion to {}, leading to division by zero.'.format(dtype))
331
    if mean.ndim == 1:
332
        mean = mean.view(-1, 1, 1)
333
    if std.ndim == 1:
334
        std = std.view(-1, 1, 1)
335
    tensor.sub_(mean).div_(std)
336
    return tensor
337
338


339
def resize(img: Tensor, size: List[int], interpolation: InterpolationMode = InterpolationMode.BILINEAR,
340
           max_size: Optional[int] = None, antialias: Optional[bool] = None) -> Tensor:
vfdev's avatar
vfdev committed
341
    r"""Resize the input image to the given size.
342
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
343
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
344

345
346
347
348
349
350
    .. 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
        types.

351
    Args:
vfdev's avatar
vfdev committed
352
        img (PIL Image or Tensor): Image to be resized.
353
354
        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
355
            the smaller edge of the image will be matched to this number maintaining
356
            the aspect ratio. i.e, if height > width, then image will be rescaled to
vfdev's avatar
vfdev committed
357
            :math:`\left(\text{size} \times \frac{\text{height}}{\text{width}}, \text{size}\right)`.
358
359
360

            .. note::
                In torchscript mode size as single int is not supported, use a sequence of length 1: ``[size, ]``.
361
362
363
364
        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.
365
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
366
367
368
369
        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
370
            ``max_size``. As a result, ``size`` might be overruled, i.e the
371
372
373
            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).
374
375
376
377
378
379
        antialias (bool, optional): antialias flag. If ``img`` is PIL Image, the flag is ignored and anti-alias
            is always used. If ``img`` is Tensor, the flag is False by default and can be set True for
            ``InterpolationMode.BILINEAR`` only mode.

            .. warning::
                There is no autodiff support for ``antialias=True`` option with input ``img`` as Tensor.
380
381

    Returns:
vfdev's avatar
vfdev committed
382
        PIL Image or Tensor: Resized image.
383
    """
384
385
386
    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
387
388
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
389
390
391
        )
        interpolation = _interpolation_modes_from_int(interpolation)

392
393
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
394

vfdev's avatar
vfdev committed
395
    if not isinstance(img, torch.Tensor):
396
397
398
399
        if antialias is not None and not antialias:
            warnings.warn(
                "Anti-alias option is always applied for PIL Image input. Argument antialias is ignored."
            )
400
        pil_interpolation = pil_modes_mapping[interpolation]
401
        return F_pil.resize(img, size=size, interpolation=pil_interpolation, max_size=max_size)
vfdev's avatar
vfdev committed
402

403
    return F_t.resize(img, size=size, interpolation=interpolation.value, max_size=max_size, antialias=antialias)
404
405
406
407
408
409
410
411


def scale(*args, **kwargs):
    warnings.warn("The use of the transforms.Scale transform is deprecated, " +
                  "please use transforms.Resize instead.")
    return resize(*args, **kwargs)


412
413
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.
414
    If the image is torch Tensor, it is expected
415
416
417
    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
418
419

    Args:
420
        img (PIL Image or Tensor): Image to be padded.
421
422
423
        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
424
            this is the padding for the left, top, right and bottom borders respectively.
425
426
427
428

            .. note::
                In torchscript mode padding as single int is not supported, use a sequence of
                length 1: ``[padding, ]``.
429
430
431
432
433
        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.
434
435
        padding_mode (str): Type of padding. Should be: constant, edge, reflect or symmetric.
            Default is constant.
436
437
438

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

439
440
            - 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
441

442
443
444
            - 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]
445

446
447
448
            - 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]
449
450

    Returns:
451
        PIL Image or Tensor: Padded image.
452
    """
453
454
    if not isinstance(img, torch.Tensor):
        return F_pil.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
455

456
    return F_t.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
457
458


vfdev's avatar
vfdev committed
459
460
def crop(img: Tensor, top: int, left: int, height: int, width: int) -> Tensor:
    """Crop the given image at specified location and output size.
461
    If the image is torch Tensor, it is expected
462
463
    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.
464

465
    Args:
vfdev's avatar
vfdev committed
466
        img (PIL Image or Tensor): Image to be cropped. (0,0) denotes the top left corner of the image.
467
468
469
470
        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.
471

472
    Returns:
vfdev's avatar
vfdev committed
473
        PIL Image or Tensor: Cropped image.
474
475
    """

vfdev's avatar
vfdev committed
476
477
    if not isinstance(img, torch.Tensor):
        return F_pil.crop(img, top, left, height, width)
478

vfdev's avatar
vfdev committed
479
    return F_t.crop(img, top, left, height, width)
480

vfdev's avatar
vfdev committed
481
482
483

def center_crop(img: Tensor, output_size: List[int]) -> Tensor:
    """Crops the given image at the center.
484
    If the image is torch Tensor, it is expected
485
486
    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.
487

488
    Args:
vfdev's avatar
vfdev committed
489
        img (PIL Image or Tensor): Image to be cropped.
490
        output_size (sequence or int): (height, width) of the crop box. If int or sequence with single int,
vfdev's avatar
vfdev committed
491
492
            it is used for both directions.

493
    Returns:
vfdev's avatar
vfdev committed
494
        PIL Image or Tensor: Cropped image.
495
    """
496
497
    if isinstance(output_size, numbers.Number):
        output_size = (int(output_size), int(output_size))
vfdev's avatar
vfdev committed
498
499
500
501
    elif isinstance(output_size, (tuple, list)) and len(output_size) == 1:
        output_size = (output_size[0], output_size[0])

    image_width, image_height = _get_image_size(img)
502
    crop_height, crop_width = output_size
vfdev's avatar
vfdev committed
503

504
505
506
507
508
509
510
511
512
513
514
515
    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
        image_width, image_height = _get_image_size(img)
        if crop_width == image_width and crop_height == image_height:
            return img

516
517
    crop_top = int(round((image_height - crop_height) / 2.))
    crop_left = int(round((image_width - crop_width) / 2.))
518
    return crop(img, crop_top, crop_left, crop_height, crop_width)
519
520


521
def resized_crop(
522
        img: Tensor, top: int, left: int, height: int, width: int, size: List[int],
523
        interpolation: InterpolationMode = InterpolationMode.BILINEAR
524
525
) -> Tensor:
    """Crop the given image and resize it to desired size.
526
    If the image is torch Tensor, it is expected
527
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
528

529
    Notably used in :class:`~torchvision.transforms.RandomResizedCrop`.
530
531

    Args:
532
        img (PIL Image or Tensor): Image to be cropped. (0,0) denotes the top left corner of the image.
533
534
535
536
        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.
537
        size (sequence or int): Desired output size. Same semantics as ``resize``.
538
539
540
541
        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.
542
543
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.

544
    Returns:
545
        PIL Image or Tensor: Cropped image.
546
    """
547
    img = crop(img, top, left, height, width)
548
549
550
551
    img = resize(img, size, interpolation)
    return img


552
def hflip(img: Tensor) -> Tensor:
553
    """Horizontally flip the given image.
554
555

    Args:
vfdev's avatar
vfdev committed
556
        img (PIL Image or Tensor): Image to be flipped. If img
557
            is a Tensor, it is expected to be in [..., H, W] format,
558
            where ... means it can have an arbitrary number of leading
559
            dimensions.
560
561

    Returns:
vfdev's avatar
vfdev committed
562
        PIL Image or Tensor:  Horizontally flipped image.
563
    """
564
565
    if not isinstance(img, torch.Tensor):
        return F_pil.hflip(img)
566

567
    return F_t.hflip(img)
568
569


570
571
572
def _get_perspective_coeffs(
        startpoints: List[List[int]], endpoints: List[List[int]]
) -> List[float]:
573
574
    """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
575
    In Perspective Transform each pixel (x, y) in the original image gets transformed as,
576
577
578
     (x, y) -> ( (ax + by + c) / (gx + hy + 1), (dx + ey + f) / (gx + hy + 1) )

    Args:
579
580
581
582
583
        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.

584
585
586
    Returns:
        octuple (a, b, c, d, e, f, g, h) for transforming each pixel.
    """
587
588
589
590
591
    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]])
592

593
    b_matrix = torch.tensor(startpoints, dtype=torch.float).view(8)
594
    res = torch.linalg.lstsq(a_matrix, b_matrix, driver='gels').solution
595

596
    output: List[float] = res.tolist()
597
    return output
598
599


600
601
602
603
def perspective(
        img: Tensor,
        startpoints: List[List[int]],
        endpoints: List[List[int]],
604
        interpolation: InterpolationMode = InterpolationMode.BILINEAR,
605
        fill: Optional[List[float]] = None
606
607
) -> Tensor:
    """Perform perspective transform of the given image.
608
    If the image is torch Tensor, it is expected
609
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
610
611

    Args:
612
613
614
615
616
        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.
617
618
619
        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.
620
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
621
622
        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.
623
624
625
626

            .. note::
                In torchscript mode single int/float value is not supported, please use a sequence
                of length 1: ``[value, ]``.
627

628
    Returns:
629
        PIL Image or Tensor: transformed Image.
630
    """
631

632
    coeffs = _get_perspective_coeffs(startpoints, endpoints)
633

634
635
636
    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
637
638
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
639
640
641
        )
        interpolation = _interpolation_modes_from_int(interpolation)

642
643
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
644

645
    if not isinstance(img, torch.Tensor):
646
647
        pil_interpolation = pil_modes_mapping[interpolation]
        return F_pil.perspective(img, coeffs, interpolation=pil_interpolation, fill=fill)
648

649
    return F_t.perspective(img, coeffs, interpolation=interpolation.value, fill=fill)
650
651


652
def vflip(img: Tensor) -> Tensor:
653
    """Vertically flip the given image.
654
655

    Args:
vfdev's avatar
vfdev committed
656
        img (PIL Image or Tensor): Image to be flipped. If img
657
            is a Tensor, it is expected to be in [..., H, W] format,
658
            where ... means it can have an arbitrary number of leading
659
            dimensions.
660
661

    Returns:
662
        PIL Image or Tensor:  Vertically flipped image.
663
    """
664
665
    if not isinstance(img, torch.Tensor):
        return F_pil.vflip(img)
666

667
    return F_t.vflip(img)
668
669


vfdev's avatar
vfdev committed
670
671
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.
672
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
673
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
674
675
676
677
678
679

    .. 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
680
681
682
        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
683
            made. If provided a sequence of length 1, it will be interpreted as (size[0], size[0]).
684

685
    Returns:
686
       tuple: tuple (tl, tr, bl, br, center)
687
       Corresponding top left, top right, bottom left, bottom right and center crop.
688
689
690
    """
    if isinstance(size, numbers.Number):
        size = (int(size), int(size))
vfdev's avatar
vfdev committed
691
692
    elif isinstance(size, (tuple, list)) and len(size) == 1:
        size = (size[0], size[0])
693

vfdev's avatar
vfdev committed
694
695
696
697
    if len(size) != 2:
        raise ValueError("Please provide only two dimensions (h, w) for size.")

    image_width, image_height = _get_image_size(img)
698
699
700
701
702
    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
703
704
705
706
707
708
709
710
    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
711
712


vfdev's avatar
vfdev committed
713
714
715
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
716
    flipped version of these (horizontal flipping is used by default).
717
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
718
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
719
720
721
722
723

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

724
    Args:
vfdev's avatar
vfdev committed
725
        img (PIL Image or Tensor): Image to be cropped.
726
        size (sequence or int): Desired output size of the crop. If size is an
727
            int instead of sequence like (h, w), a square crop (size, size) is
728
            made. If provided a sequence of length 1, it will be interpreted as (size[0], size[0]).
729
        vertical_flip (bool): Use vertical flipping instead of horizontal
730
731

    Returns:
732
        tuple: tuple (tl, tr, bl, br, center, tl_flip, tr_flip, bl_flip, br_flip, center_flip)
733
734
        Corresponding top left, top right, bottom left, bottom right and
        center crop and same for the flipped image.
735
736
737
    """
    if isinstance(size, numbers.Number):
        size = (int(size), int(size))
vfdev's avatar
vfdev committed
738
739
740
741
742
    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.")
743
744
745
746
747
748
749
750
751
752
753
754

    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


755
def adjust_brightness(img: Tensor, brightness_factor: float) -> Tensor:
756
    """Adjust brightness of an image.
757
758

    Args:
vfdev's avatar
vfdev committed
759
        img (PIL Image or Tensor): Image to be adjusted.
760
761
            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.
762
763
764
765
766
        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
767
        PIL Image or Tensor: Brightness adjusted image.
768
    """
769
770
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_brightness(img, brightness_factor)
771

772
    return F_t.adjust_brightness(img, brightness_factor)
773
774


775
def adjust_contrast(img: Tensor, contrast_factor: float) -> Tensor:
776
    """Adjust contrast of an image.
777
778

    Args:
vfdev's avatar
vfdev committed
779
        img (PIL Image or Tensor): Image to be adjusted.
780
781
            If img is torch Tensor, it is expected to be in [..., 3, H, W] format,
            where ... means it can have an arbitrary number of leading dimensions.
782
783
784
785
786
        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
787
        PIL Image or Tensor: Contrast adjusted image.
788
    """
789
790
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_contrast(img, contrast_factor)
791

792
    return F_t.adjust_contrast(img, contrast_factor)
793
794


795
def adjust_saturation(img: Tensor, saturation_factor: float) -> Tensor:
796
797
798
    """Adjust color saturation of an image.

    Args:
vfdev's avatar
vfdev committed
799
        img (PIL Image or Tensor): Image to be adjusted.
800
801
            If img is torch Tensor, it is expected to be in [..., 3, H, W] format,
            where ... means it can have an arbitrary number of leading dimensions.
802
803
804
805
806
        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
807
        PIL Image or Tensor: Saturation adjusted image.
808
    """
809
810
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_saturation(img, saturation_factor)
811

812
    return F_t.adjust_saturation(img, saturation_factor)
813
814


815
def adjust_hue(img: Tensor, hue_factor: float) -> Tensor:
816
817
818
819
820
821
822
823
824
    """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]`.

825
826
827
    See `Hue`_ for more details.

    .. _Hue: https://en.wikipedia.org/wiki/Hue
828
829

    Args:
830
        img (PIL Image or Tensor): Image to be adjusted.
831
832
833
            If img is torch Tensor, it is expected to be in [..., 3, H, W] format,
            where ... means it can have an arbitrary number of leading dimensions.
            If img is PIL Image mode "1", "L", "I", "F" and modes with transparency (alpha channel) are not supported.
834
835
836
837
838
839
840
        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:
841
        PIL Image or Tensor: Hue adjusted image.
842
    """
843
844
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_hue(img, hue_factor)
845

846
    return F_t.adjust_hue(img, hue_factor)
847
848


849
def adjust_gamma(img: Tensor, gamma: float, gain: float = 1) -> Tensor:
850
    r"""Perform gamma correction on an image.
851
852
853
854

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

855
856
857
858
    .. math::
        I_{\text{out}} = 255 \times \text{gain} \times \left(\frac{I_{\text{in}}}{255}\right)^{\gamma}

    See `Gamma Correction`_ for more details.
859

860
    .. _Gamma Correction: https://en.wikipedia.org/wiki/Gamma_correction
861
862

    Args:
863
        img (PIL Image or Tensor): PIL Image to be adjusted.
864
865
866
            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.
867
868
869
        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.
870
        gain (float): The constant multiplier.
871
872
    Returns:
        PIL Image or Tensor: Gamma correction adjusted image.
873
    """
874
875
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_gamma(img, gamma, gain)
876

877
    return F_t.adjust_gamma(img, gamma, gain)
878
879


vfdev's avatar
vfdev committed
880
def _get_inverse_affine_matrix(
vfdev's avatar
vfdev committed
881
        center: List[float], angle: float, translate: List[float], scale: float, shear: List[float]
vfdev's avatar
vfdev committed
882
) -> List[float]:
883
884
885
886
887
888
889
    # Helper method to compute inverse matrix for affine transformation

    # As it is explained in PIL.Image.rotate
    # We need compute INVERSE of affine transformation matrix: M = T * C * RSS * C^-1
    # 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]
    #       RSS is rotation with scale and shear matrix
890
891
892
893
894
895
896
897
898
899
    #       RSS(a, s, (sx, sy)) =
    #       = R(a) * S(s) * SHy(sy) * SHx(sx)
    #       = [ s*cos(a - sy)/cos(sy), s*(-cos(a - sy)*tan(x)/cos(y) - sin(a)), 0 ]
    #         [ s*sin(a + sy)/cos(sy), s*(-sin(a - sy)*tan(x)/cos(y) + cos(a)), 0 ]
    #         [ 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]
    #
900
901
    # Thus, the inverse is M^-1 = C * RSS^-1 * C^-1 * T^-1

902
903
904
905
906
907
908
    rot = math.radians(angle)
    sx, sy = [math.radians(s) for s in shear]

    cx, cy = center
    tx, ty = translate

    # RSS without scaling
vfdev's avatar
vfdev committed
909
910
911
912
    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)
913
914

    # Inverted rotation matrix with scale and shear
915
    # det([[a, b], [c, d]]) == 1, since det(rotation) = 1 and det(shear) = 1
vfdev's avatar
vfdev committed
916
917
    matrix = [d, -b, 0.0, -c, a, 0.0]
    matrix = [x / scale for x in matrix]
918
919

    # Apply inverse of translation and of center translation: RSS^-1 * C^-1 * T^-1
vfdev's avatar
vfdev committed
920
921
    matrix[2] += matrix[0] * (-cx - tx) + matrix[1] * (-cy - ty)
    matrix[5] += matrix[3] * (-cx - tx) + matrix[4] * (-cy - ty)
922
923

    # Apply center translation: C * RSS^-1 * C^-1 * T^-1
vfdev's avatar
vfdev committed
924
925
    matrix[2] += cx
    matrix[5] += cy
926

vfdev's avatar
vfdev committed
927
    return matrix
928

vfdev's avatar
vfdev committed
929

vfdev's avatar
vfdev committed
930
def rotate(
931
        img: Tensor, angle: float, interpolation: InterpolationMode = InterpolationMode.NEAREST,
932
        expand: bool = False, center: Optional[List[int]] = None,
933
        fill: Optional[List[float]] = None, resample: Optional[int] = None
vfdev's avatar
vfdev committed
934
935
) -> Tensor:
    """Rotate the image by angle.
936
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
937
938
939
940
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.

    Args:
        img (PIL Image or Tensor): image to be rotated.
941
        angle (number): rotation angle value in degrees, counter-clockwise.
942
943
944
        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.
945
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
vfdev's avatar
vfdev committed
946
947
948
949
        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.
950
        center (sequence, optional): Optional center of rotation. Origin is the upper left corner.
vfdev's avatar
vfdev committed
951
            Default is the center of the image.
952
953
        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.
954
955
956
957

            .. note::
                In torchscript mode single int/float value is not supported, please use a sequence
                of length 1: ``[value, ]``.
vfdev's avatar
vfdev committed
958
959
960
961
962
963
964

    Returns:
        PIL Image or Tensor: Rotated image.

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

    """
965
966
967
968
969
970
971
972
973
    if resample is not None:
        warnings.warn(
            "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
        )
        interpolation = _interpolation_modes_from_int(resample)

    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
974
975
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
976
977
978
        )
        interpolation = _interpolation_modes_from_int(interpolation)

vfdev's avatar
vfdev committed
979
980
981
982
983
984
    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")

985
986
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
987

vfdev's avatar
vfdev committed
988
    if not isinstance(img, torch.Tensor):
989
990
        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
991
992
993
994

    center_f = [0.0, 0.0]
    if center is not None:
        img_size = _get_image_size(img)
995
996
997
        # Center values should be in pixel coordinates but translated such that (0, 0) corresponds to image center.
        center_f = [1.0 * (c - s * 0.5) for c, s in zip(center, img_size)]

vfdev's avatar
vfdev committed
998
999
1000
    # 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])
1001
    return F_t.rotate(img, matrix=matrix, interpolation=interpolation.value, expand=expand, fill=fill)
vfdev's avatar
vfdev committed
1002
1003


vfdev's avatar
vfdev committed
1004
1005
def affine(
        img: Tensor, angle: float, translate: List[int], scale: float, shear: List[float],
1006
1007
        interpolation: InterpolationMode = InterpolationMode.NEAREST, fill: Optional[List[float]] = None,
        resample: Optional[int] = None, fillcolor: Optional[List[float]] = None
vfdev's avatar
vfdev committed
1008
1009
) -> Tensor:
    """Apply affine transformation on the image keeping image center invariant.
1010
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
1011
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
1012
1013

    Args:
vfdev's avatar
vfdev committed
1014
        img (PIL Image or Tensor): image to transform.
1015
1016
        angle (number): rotation angle in degrees between -180 and 180, clockwise direction.
        translate (sequence of integers): horizontal and vertical translations (post-rotation translation)
1017
        scale (float): overall scale
1018
1019
        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
1020
            the second value corresponds to a shear parallel to the y axis.
1021
1022
1023
        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.
1024
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
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
        fillcolor (sequence, int, float): deprecated argument and will be removed since v0.10.0.
1032
            Please use the ``fill`` parameter instead.
1033
        resample (int, optional): deprecated argument and will be removed since v0.10.0.
1034
            Please use the ``interpolation`` parameter instead.
vfdev's avatar
vfdev committed
1035
1036
1037

    Returns:
        PIL Image or Tensor: Transformed image.
1038
    """
1039
1040
1041
1042
1043
1044
1045
1046
1047
    if resample is not None:
        warnings.warn(
            "Argument resample is deprecated and will be removed since v0.10.0. Please, use interpolation instead"
        )
        interpolation = _interpolation_modes_from_int(resample)

    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
1048
1049
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
1050
1051
1052
1053
1054
1055
1056
1057
1058
        )
        interpolation = _interpolation_modes_from_int(interpolation)

    if fillcolor is not None:
        warnings.warn(
            "Argument fillcolor is deprecated and will be removed since v0.10.0. Please, use fill instead"
        )
        fill = fillcolor

vfdev's avatar
vfdev committed
1059
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
    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")

1074
1075
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
1076

vfdev's avatar
vfdev committed
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
    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:
        raise ValueError("Shear should be a sequence containing two values. Got {}".format(shear))

    img_size = _get_image_size(img)
    if not isinstance(img, torch.Tensor):
        # center = (img_size[0] * 0.5 + 0.5, img_size[1] * 0.5 + 0.5)
        # 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
        center = [img_size[0] * 0.5, img_size[1] * 0.5]
        matrix = _get_inverse_affine_matrix(center, angle, translate, scale, shear)
1102
1103
        pil_interpolation = pil_modes_mapping[interpolation]
        return F_pil.affine(img, matrix=matrix, interpolation=pil_interpolation, fill=fill)
1104

1105
1106
    translate_f = [1.0 * t for t in translate]
    matrix = _get_inverse_affine_matrix([0.0, 0.0], angle, translate_f, scale, shear)
1107
    return F_t.affine(img, matrix=matrix, interpolation=interpolation.value, fill=fill)
1108
1109


1110
@torch.jit.unused
1111
def to_grayscale(img, num_output_channels=1):
1112
    """Convert PIL image of any mode (RGB, HSV, LAB, etc) to grayscale version of image.
1113
    This transform does not support torch Tensor.
1114
1115

    Args:
1116
        img (PIL Image): PIL Image to be converted to grayscale.
1117
        num_output_channels (int): number of channels of the output image. Value can be 1 or 3. Default is 1.
1118
1119

    Returns:
1120
1121
        PIL Image: Grayscale version of the image.

1122
1123
        - if num_output_channels = 1 : returned image is single channel
        - if num_output_channels = 3 : returned image is 3 channel with r = g = b
1124
    """
1125
1126
    if isinstance(img, Image.Image):
        return F_pil.to_grayscale(img, num_output_channels)
1127

1128
1129
1130
1131
1132
    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.
1133
1134
    If the image is torch Tensor, it is expected
    to have [..., 3, H, W] shape, where ... means an arbitrary number of leading dimensions
1135
1136
1137
1138
1139
1140
1141
1142
1143
1144
1145
1146

    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.

1147
1148
        - if num_output_channels = 1 : returned image is single channel
        - if num_output_channels = 3 : returned image is 3 channel with r = g = b
1149
1150
1151
1152
1153
    """
    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)
1154
1155


1156
def erase(img: Tensor, i: int, j: int, h: int, w: int, v: Tensor, inplace: bool = False) -> Tensor:
1157
    """ Erase the input Tensor Image with given value.
1158
    This transform does not support PIL Image.
1159
1160
1161
1162
1163
1164
1165
1166

    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
1167
        inplace(bool, optional): For in-place operations. By default is set False.
1168
1169
1170
1171
1172
1173
1174

    Returns:
        Tensor Image: Erased image.
    """
    if not isinstance(img, torch.Tensor):
        raise TypeError('img should be Tensor Image. Got {}'.format(type(img)))

1175
1176
1177
    if not inplace:
        img = img.clone()

vfdev's avatar
vfdev committed
1178
    img[..., i:i + h, j:j + w] = v
1179
    return img
1180
1181
1182


def gaussian_blur(img: Tensor, kernel_size: List[int], sigma: Optional[List[float]] = None) -> Tensor:
1183
1184
1185
    """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.
1186
1187
1188
1189
1190

    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.
1191
1192
1193
1194

            .. note::
                In torchscript mode kernel_size as single int is not supported, use a sequence of
                length 1: ``[ksize, ]``.
1195
1196
1197
1198
        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``.
1199
1200
1201
1202
1203
            Default, None.

            .. note::
                In torchscript mode sigma as single float is
                not supported, use a sequence of length 1: ``[sigma, ]``.
1204
1205
1206
1207
1208
1209
1210
1211
1212
1213
1214
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
1228
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244

    Returns:
        PIL Image or Tensor: Gaussian Blurred version of the image.
    """
    if not isinstance(kernel_size, (int, list, tuple)):
        raise TypeError('kernel_size should be int or a sequence of integers. Got {}'.format(type(kernel_size)))
    if isinstance(kernel_size, int):
        kernel_size = [kernel_size, kernel_size]
    if len(kernel_size) != 2:
        raise ValueError('If kernel_size is a sequence its length should be 2. Got {}'.format(len(kernel_size)))
    for ksize in kernel_size:
        if ksize % 2 == 0 or ksize < 0:
            raise ValueError('kernel_size should have odd and positive integers. Got {}'.format(kernel_size))

    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)):
        raise TypeError('sigma should be either float or sequence of floats. Got {}'.format(type(sigma)))
    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:
        raise ValueError('If sigma is a sequence, its length should be 2. Got {}'.format(len(sigma)))
    for s in sigma:
        if s <= 0.:
            raise ValueError('sigma should have positive values. Got {}'.format(sigma))

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

        t_img = to_tensor(img)

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

    if not isinstance(img, torch.Tensor):
        output = to_pil_image(output)
    return output
1245
1246
1247


def invert(img: Tensor) -> Tensor:
1248
    """Invert the colors of an RGB/grayscale image.
1249
1250
1251

    Args:
        img (PIL Image or Tensor): Image to have its colors inverted.
1252
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1253
1254
            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".
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265

    Returns:
        PIL Image or Tensor: Color inverted image.
    """
    if not isinstance(img, torch.Tensor):
        return F_pil.invert(img)

    return F_t.invert(img)


def posterize(img: Tensor, bits: int) -> Tensor:
1266
    """Posterize an image by reducing the number of bits for each color channel.
1267
1268
1269

    Args:
        img (PIL Image or Tensor): Image to have its colors posterized.
1270
            If img is torch Tensor, it should be of type torch.uint8 and
1271
1272
1273
            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".
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
        bits (int): The number of bits to keep for each channel (0-8).
    Returns:
        PIL Image or Tensor: Posterized image.
    """
    if not (0 <= bits <= 8):
        raise ValueError('The number if bits should be between 0 and 8. Got {}'.format(bits))

    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:
1288
    """Solarize an RGB/grayscale image by inverting all pixel values above a threshold.
1289
1290
1291

    Args:
        img (PIL Image or Tensor): Image to have its colors inverted.
1292
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1293
1294
            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".
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
        threshold (float): All pixels equal or above this value are inverted.
    Returns:
        PIL Image or Tensor: Solarized image.
    """
    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:
1306
    """Adjust the sharpness of an image.
1307
1308
1309

    Args:
        img (PIL Image or Tensor): Image to be adjusted.
1310
1311
            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.
1312
1313
1314
1315
1316
1317
1318
1319
1320
1321
1322
1323
1324
1325
        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.
    """
    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:
1326
    """Maximize contrast of an image by remapping its
1327
1328
1329
1330
1331
    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.
1332
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1333
1334
            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".
1335
1336
1337
1338
1339
1340
1341
1342
1343
1344
1345

    Returns:
        PIL Image or Tensor: An image that was autocontrasted.
    """
    if not isinstance(img, torch.Tensor):
        return F_pil.autocontrast(img)

    return F_t.autocontrast(img)


def equalize(img: Tensor) -> Tensor:
1346
    """Equalize the histogram of an image by applying
1347
1348
1349
1350
1351
    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.
1352
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1353
            where ... means it can have an arbitrary number of leading dimensions.
1354
            The tensor dtype must be ``torch.uint8`` and values are expected to be in ``[0, 255]``.
1355
            If img is PIL Image, it is expected to be in mode "P", "L" or "RGB".
1356
1357
1358
1359
1360
1361
1362
1363

    Returns:
        PIL Image or Tensor: An image that was equalized.
    """
    if not isinstance(img, torch.Tensor):
        return F_pil.equalize(img)

    return F_t.equalize(img)