functional.py 56.4 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
18
19
from . import functional_pil as F_pil
from . import functional_tensor as F_t

20

21
class InterpolationMode(Enum):
22
    """Interpolation modes
23
    Available interpolation methods are ``nearest``, ``bilinear``, ``bicubic``, ``box``, ``hamming``, and ``lanczos``.
24
    """
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
_is_pil_image = F_pil._is_pil_image


61
62
63
64
65
66
67
68
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
69
70
    """
    if isinstance(img, torch.Tensor):
71
        return F_t.get_image_size(img)
72

73
    return F_pil.get_image_size(img)
74

vfdev's avatar
vfdev committed
75

76
77
78
79
80
81
82
83
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.
84
    """
85
    if isinstance(img, torch.Tensor):
86
        return F_t.get_image_num_channels(img)
87

88
    return F_pil.get_image_num_channels(img)
89
90


vfdev's avatar
vfdev committed
91
92
@torch.jit.unused
def _is_numpy(img: Any) -> bool:
93
94
95
    return isinstance(img, np.ndarray)


vfdev's avatar
vfdev committed
96
97
@torch.jit.unused
def _is_numpy_image(img: Any) -> bool:
98
    return img.ndim in {2, 3}
99
100
101
102


def to_tensor(pic):
    """Convert a ``PIL Image`` or ``numpy.ndarray`` to tensor.
103
    This function does not support torchscript.
104

105
    See :class:`~torchvision.transforms.ToTensor` for more details.
106
107
108
109
110
111
112

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

    Returns:
        Tensor: Converted image.
    """
113
    if not (F_pil._is_pil_image(pic) or _is_numpy(pic)):
114
        raise TypeError(f"pic should be PIL Image or ndarray. Got {type(pic)}")
115

116
    if _is_numpy(pic) and not _is_numpy_image(pic):
117
        raise ValueError(f"pic should be 2/3 dimensional. Got {pic.ndim} dimensions.")
118

119
120
    default_float_dtype = torch.get_default_dtype()

121
122
    if isinstance(pic, np.ndarray):
        # handle numpy array
surgan12's avatar
surgan12 committed
123
124
125
        if pic.ndim == 2:
            pic = pic[:, :, None]

126
        img = torch.from_numpy(pic.transpose((2, 0, 1))).contiguous()
127
        # backward compatibility
128
        if isinstance(img, torch.ByteTensor):
129
            return img.to(dtype=default_float_dtype).div(255)
130
131
        else:
            return img
132
133

    if accimage is not None and isinstance(pic, accimage.Image):
134
        nppic = np.zeros([pic.channels, pic.height, pic.width], dtype=np.float32)
135
        pic.copyto(nppic)
136
        return torch.from_numpy(nppic).to(dtype=default_float_dtype)
137
138

    # handle PIL Image
139
140
    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))
141

142
    if pic.mode == "1":
143
        img = 255 * img
144
    img = img.view(pic.size[1], pic.size[0], len(pic.getbands()))
145
    # put it from HWC to CHW format
146
    img = img.permute((2, 0, 1)).contiguous()
147
    if isinstance(img, torch.ByteTensor):
148
        return img.to(dtype=default_float_dtype).div(255)
149
150
151
152
    else:
        return img


153
154
def pil_to_tensor(pic):
    """Convert a ``PIL Image`` to a tensor of the same type.
155
    This function does not support torchscript.
156

vfdev's avatar
vfdev committed
157
    See :class:`~torchvision.transforms.PILToTensor` for more details.
158

159
160
161
162
    .. note::

        A deep copy of the underlying array is performed.

163
164
165
166
167
168
    Args:
        pic (PIL Image): Image to be converted to tensor.

    Returns:
        Tensor: Converted image.
    """
169
    if not F_pil._is_pil_image(pic):
170
        raise TypeError(f"pic should be PIL Image. Got {type(pic)}")
171
172

    if accimage is not None and isinstance(pic, accimage.Image):
173
174
        # accimage format is always uint8 internally, so always return uint8 here
        nppic = np.zeros([pic.channels, pic.height, pic.width], dtype=np.uint8)
175
176
177
178
        pic.copyto(nppic)
        return torch.as_tensor(nppic)

    # handle PIL Image
179
    img = torch.as_tensor(np.array(pic, copy=True))
180
181
182
183
184
185
    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


186
187
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
188
    This function does not support PIL Image.
189
190
191
192
193
194

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

    Returns:
vfdev's avatar
vfdev committed
195
        Tensor: Converted image
196
197
198
199
200
201
202
203
204
205
206
207

    .. 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``.
    """
208
    if not isinstance(image, torch.Tensor):
209
        raise TypeError("Input img should be Tensor Image")
210
211

    return F_t.convert_image_dtype(image, dtype)
212
213


214
def to_pil_image(pic, mode=None):
215
    """Convert a tensor or an ndarray to PIL Image. This function does not support torchscript.
216

217
    See :class:`~torchvision.transforms.ToPILImage` for more details.
218
219
220
221
222

    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).

223
    .. _PIL.Image mode: https://pillow.readthedocs.io/en/latest/handbook/concepts.html#concept-modes
224
225
226
227

    Returns:
        PIL Image: Image converted to PIL Image.
    """
228
    if not (isinstance(pic, torch.Tensor) or isinstance(pic, np.ndarray)):
229
        raise TypeError(f"pic should be Tensor or ndarray. Got {type(pic)}.")
230

Varun Agrawal's avatar
Varun Agrawal committed
231
232
    elif isinstance(pic, torch.Tensor):
        if pic.ndimension() not in {2, 3}:
233
            raise ValueError(f"pic should be 2/3 dimensional. Got {pic.ndimension()} dimensions.")
Varun Agrawal's avatar
Varun Agrawal committed
234
235
236

        elif pic.ndimension() == 2:
            # if 2D image, add channel dimension (CHW)
Surgan Jandial's avatar
Surgan Jandial committed
237
            pic = pic.unsqueeze(0)
Varun Agrawal's avatar
Varun Agrawal committed
238

239
240
        # check number of channels
        if pic.shape[-3] > 4:
241
            raise ValueError(f"pic should not have > 4 channels. Got {pic.shape[-3]} channels.")
242

Varun Agrawal's avatar
Varun Agrawal committed
243
244
    elif isinstance(pic, np.ndarray):
        if pic.ndim not in {2, 3}:
245
            raise ValueError(f"pic should be 2/3 dimensional. Got {pic.ndim} dimensions.")
Varun Agrawal's avatar
Varun Agrawal committed
246
247
248
249
250

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

251
252
        # check number of channels
        if pic.shape[-1] > 4:
253
            raise ValueError(f"pic should not have > 4 channels. Got {pic.shape[-1]} channels.")
254

255
    npimg = pic
Varun Agrawal's avatar
Varun Agrawal committed
256
    if isinstance(pic, torch.Tensor):
257
        if pic.is_floating_point() and mode != "F":
258
259
            pic = pic.mul(255).byte()
        npimg = np.transpose(pic.cpu().numpy(), (1, 2, 0))
260
261

    if not isinstance(npimg, np.ndarray):
262
        raise TypeError("Input pic must be a torch.Tensor or NumPy ndarray, not {type(npimg)}")
263
264
265
266
267

    if npimg.shape[2] == 1:
        expected_mode = None
        npimg = npimg[:, :, 0]
        if npimg.dtype == np.uint8:
268
            expected_mode = "L"
vfdev's avatar
vfdev committed
269
        elif npimg.dtype == np.int16:
270
            expected_mode = "I;16"
vfdev's avatar
vfdev committed
271
        elif npimg.dtype == np.int32:
272
            expected_mode = "I"
273
        elif npimg.dtype == np.float32:
274
            expected_mode = "F"
275
        if mode is not None and mode != expected_mode:
276
            raise ValueError(f"Incorrect mode ({mode}) supplied for input type {np.dtype}. Should be {expected_mode}")
277
278
        mode = expected_mode

surgan12's avatar
surgan12 committed
279
    elif npimg.shape[2] == 2:
280
        permitted_2_channel_modes = ["LA"]
surgan12's avatar
surgan12 committed
281
        if mode is not None and mode not in permitted_2_channel_modes:
282
            raise ValueError(f"Only modes {permitted_2_channel_modes} are supported for 2D inputs")
surgan12's avatar
surgan12 committed
283
284

        if mode is None and npimg.dtype == np.uint8:
285
            mode = "LA"
surgan12's avatar
surgan12 committed
286

287
    elif npimg.shape[2] == 4:
288
        permitted_4_channel_modes = ["RGBA", "CMYK", "RGBX"]
289
        if mode is not None and mode not in permitted_4_channel_modes:
290
            raise ValueError(f"Only modes {permitted_4_channel_modes} are supported for 4D inputs")
291
292

        if mode is None and npimg.dtype == np.uint8:
293
            mode = "RGBA"
294
    else:
295
        permitted_3_channel_modes = ["RGB", "YCbCr", "HSV"]
296
        if mode is not None and mode not in permitted_3_channel_modes:
297
            raise ValueError(f"Only modes {permitted_3_channel_modes} are supported for 3D inputs")
298
        if mode is None and npimg.dtype == np.uint8:
299
            mode = "RGB"
300
301

    if mode is None:
302
        raise TypeError(f"Input type {npimg.dtype} is not supported")
303
304
305
306

    return Image.fromarray(npimg, mode=mode)


307
def normalize(tensor: Tensor, mean: List[float], std: List[float], inplace: bool = False) -> Tensor:
308
    """Normalize a float tensor image with mean and standard deviation.
309
    This transform does not support PIL Image.
310

311
    .. note::
surgan12's avatar
surgan12 committed
312
        This transform acts out of place by default, i.e., it does not mutates the input tensor.
313

314
    See :class:`~torchvision.transforms.Normalize` for more details.
315
316

    Args:
317
        tensor (Tensor): Float tensor image of size (C, H, W) or (B, C, H, W) to be normalized.
318
        mean (sequence): Sequence of means for each channel.
319
        std (sequence): Sequence of standard deviations for each channel.
320
        inplace(bool,optional): Bool to make this operation inplace.
321
322
323
324

    Returns:
        Tensor: Normalized Tensor image.
    """
325
    if not isinstance(tensor, torch.Tensor):
326
        raise TypeError(f"Input tensor should be a torch tensor. Got {type(tensor)}.")
327

328
    if not tensor.is_floating_point():
329
        raise TypeError(f"Input tensor should be a float tensor. Got {tensor.dtype}.")
330

331
    if tensor.ndim < 3:
332
        raise ValueError(
333
            f"Expected tensor to be a tensor image of size (..., C, H, W). Got tensor.size() = {tensor.size()}"
334
        )
335

surgan12's avatar
surgan12 committed
336
337
338
    if not inplace:
        tensor = tensor.clone()

339
340
341
    dtype = tensor.dtype
    mean = torch.as_tensor(mean, dtype=dtype, device=tensor.device)
    std = torch.as_tensor(std, dtype=dtype, device=tensor.device)
342
    if (std == 0).any():
343
        raise ValueError(f"std evaluated to zero after conversion to {dtype}, leading to division by zero.")
344
    if mean.ndim == 1:
345
        mean = mean.view(-1, 1, 1)
346
    if std.ndim == 1:
347
        std = std.view(-1, 1, 1)
348
    tensor.sub_(mean).div_(std)
349
    return tensor
350
351


352
353
354
355
356
357
358
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
359
    r"""Resize the input image to the given size.
360
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
361
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
362

363
364
365
366
    .. 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
367
368
        types. See also below the ``antialias`` parameter, which can help making the output of PIL images and tensors
        closer.
369

370
    Args:
vfdev's avatar
vfdev committed
371
        img (PIL Image or Tensor): Image to be resized.
372
373
        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
374
            the smaller edge of the image will be matched to this number maintaining
375
            the aspect ratio. i.e, if height > width, then image will be rescaled to
vfdev's avatar
vfdev committed
376
            :math:`\left(\text{size} \times \frac{\text{height}}{\text{width}}, \text{size}\right)`.
377
378
379

            .. note::
                In torchscript mode size as single int is not supported, use a sequence of length 1: ``[size, ]``.
380
381
382
383
        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.
384
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
385
386
387
388
        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
389
            ``max_size``. As a result, ``size`` might be overruled, i.e the
390
391
392
            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).
393
        antialias (bool, optional): antialias flag. If ``img`` is PIL Image, the flag is ignored and anti-alias
394
395
396
            is always used. If ``img`` is Tensor, the flag is False by default and can be set to True for
            ``InterpolationMode.BILINEAR`` only mode. This can help making the output for PIL images and tensors
            closer.
397
398
399

            .. warning::
                There is no autodiff support for ``antialias=True`` option with input ``img`` as Tensor.
400
401

    Returns:
vfdev's avatar
vfdev committed
402
        PIL Image or Tensor: Resized image.
403
    """
404
405
406
    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
407
408
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
409
410
411
        )
        interpolation = _interpolation_modes_from_int(interpolation)

412
413
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
414

vfdev's avatar
vfdev committed
415
    if not isinstance(img, torch.Tensor):
416
        if antialias is not None and not antialias:
417
            warnings.warn("Anti-alias option is always applied for PIL Image input. Argument antialias is ignored.")
418
        pil_interpolation = pil_modes_mapping[interpolation]
419
        return F_pil.resize(img, size=size, interpolation=pil_interpolation, max_size=max_size)
vfdev's avatar
vfdev committed
420

421
    return F_t.resize(img, size=size, interpolation=interpolation.value, max_size=max_size, antialias=antialias)
422
423
424


def scale(*args, **kwargs):
425
    warnings.warn("The use of the transforms.Scale transform is deprecated, please use transforms.Resize instead.")
426
427
428
    return resize(*args, **kwargs)


429
430
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.
431
    If the image is torch Tensor, it is expected
432
433
434
    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
435
436

    Args:
437
        img (PIL Image or Tensor): Image to be padded.
438
439
440
        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
441
            this is the padding for the left, top, right and bottom borders respectively.
442
443
444
445

            .. note::
                In torchscript mode padding as single int is not supported, use a sequence of
                length 1: ``[padding, ]``.
446
447
448
449
450
        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.
451
452
        padding_mode (str): Type of padding. Should be: constant, edge, reflect or symmetric.
            Default is constant.
453
454
455

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

456
457
            - 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
458

459
460
461
            - 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]
462

463
464
465
            - 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]
466
467

    Returns:
468
        PIL Image or Tensor: Padded image.
469
    """
470
471
    if not isinstance(img, torch.Tensor):
        return F_pil.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
472

473
    return F_t.pad(img, padding=padding, fill=fill, padding_mode=padding_mode)
474
475


vfdev's avatar
vfdev committed
476
477
def crop(img: Tensor, top: int, left: int, height: int, width: int) -> Tensor:
    """Crop the given image at specified location and output size.
478
    If the image is torch Tensor, it is expected
479
480
    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.
481

482
    Args:
vfdev's avatar
vfdev committed
483
        img (PIL Image or Tensor): Image to be cropped. (0,0) denotes the top left corner of the image.
484
485
486
487
        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.
488

489
    Returns:
vfdev's avatar
vfdev committed
490
        PIL Image or Tensor: Cropped image.
491
492
    """

vfdev's avatar
vfdev committed
493
494
    if not isinstance(img, torch.Tensor):
        return F_pil.crop(img, top, left, height, width)
495

vfdev's avatar
vfdev committed
496
    return F_t.crop(img, top, left, height, width)
497

vfdev's avatar
vfdev committed
498
499
500

def center_crop(img: Tensor, output_size: List[int]) -> Tensor:
    """Crops the given image at the center.
501
    If the image is torch Tensor, it is expected
502
503
    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.
504

505
    Args:
vfdev's avatar
vfdev committed
506
        img (PIL Image or Tensor): Image to be cropped.
507
        output_size (sequence or int): (height, width) of the crop box. If int or sequence with single int,
vfdev's avatar
vfdev committed
508
509
            it is used for both directions.

510
    Returns:
vfdev's avatar
vfdev committed
511
        PIL Image or Tensor: Cropped image.
512
    """
513
514
    if isinstance(output_size, numbers.Number):
        output_size = (int(output_size), int(output_size))
vfdev's avatar
vfdev committed
515
516
517
    elif isinstance(output_size, (tuple, list)) and len(output_size) == 1:
        output_size = (output_size[0], output_size[0])

518
    image_width, image_height = get_image_size(img)
519
    crop_height, crop_width = output_size
vfdev's avatar
vfdev committed
520

521
522
523
524
525
526
527
528
    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
529
        image_width, image_height = get_image_size(img)
530
531
532
        if crop_width == image_width and crop_height == image_height:
            return img

533
534
    crop_top = int(round((image_height - crop_height) / 2.0))
    crop_left = int(round((image_width - crop_width) / 2.0))
535
    return crop(img, crop_top, crop_left, crop_height, crop_width)
536
537


538
def resized_crop(
539
540
541
542
543
544
545
    img: Tensor,
    top: int,
    left: int,
    height: int,
    width: int,
    size: List[int],
    interpolation: InterpolationMode = InterpolationMode.BILINEAR,
546
547
) -> Tensor:
    """Crop the given image and resize it to desired size.
548
    If the image is torch Tensor, it is expected
549
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
550

551
    Notably used in :class:`~torchvision.transforms.RandomResizedCrop`.
552
553

    Args:
554
        img (PIL Image or Tensor): Image to be cropped. (0,0) denotes the top left corner of the image.
555
556
557
558
        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.
559
        size (sequence or int): Desired output size. Same semantics as ``resize``.
560
561
562
563
        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.
564
565
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.

566
    Returns:
567
        PIL Image or Tensor: Cropped image.
568
    """
569
    img = crop(img, top, left, height, width)
570
571
572
573
    img = resize(img, size, interpolation)
    return img


574
def hflip(img: Tensor) -> Tensor:
575
    """Horizontally flip the given image.
576
577

    Args:
vfdev's avatar
vfdev committed
578
        img (PIL Image or Tensor): Image to be flipped. If img
579
            is a Tensor, it is expected to be in [..., H, W] format,
580
            where ... means it can have an arbitrary number of leading
581
            dimensions.
582
583

    Returns:
vfdev's avatar
vfdev committed
584
        PIL Image or Tensor:  Horizontally flipped image.
585
    """
586
587
    if not isinstance(img, torch.Tensor):
        return F_pil.hflip(img)
588

589
    return F_t.hflip(img)
590
591


592
def _get_perspective_coeffs(startpoints: List[List[int]], endpoints: List[List[int]]) -> List[float]:
593
594
    """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
595
    In Perspective Transform each pixel (x, y) in the original image gets transformed as,
596
597
598
     (x, y) -> ( (ax + by + c) / (gx + hy + 1), (dx + ey + f) / (gx + hy + 1) )

    Args:
599
600
601
602
603
        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.

604
605
606
    Returns:
        octuple (a, b, c, d, e, f, g, h) for transforming each pixel.
    """
607
608
609
610
611
    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]])
612

613
    b_matrix = torch.tensor(startpoints, dtype=torch.float).view(8)
614
    res = torch.linalg.lstsq(a_matrix, b_matrix, driver="gels").solution
615

616
    output: List[float] = res.tolist()
617
    return output
618
619


620
def perspective(
621
622
623
624
625
    img: Tensor,
    startpoints: List[List[int]],
    endpoints: List[List[int]],
    interpolation: InterpolationMode = InterpolationMode.BILINEAR,
    fill: Optional[List[float]] = None,
626
627
) -> Tensor:
    """Perform perspective transform of the given image.
628
    If the image is torch Tensor, it is expected
629
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
630
631

    Args:
632
633
634
635
636
        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.
637
638
639
        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.
640
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
641
642
        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.
643
644
645
646

            .. note::
                In torchscript mode single int/float value is not supported, please use a sequence
                of length 1: ``[value, ]``.
647

648
    Returns:
649
        PIL Image or Tensor: transformed Image.
650
    """
651

652
    coeffs = _get_perspective_coeffs(startpoints, endpoints)
653

654
655
656
    # Backward compatibility with integer value
    if isinstance(interpolation, int):
        warnings.warn(
657
658
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
659
660
661
        )
        interpolation = _interpolation_modes_from_int(interpolation)

662
663
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
664

665
    if not isinstance(img, torch.Tensor):
666
667
        pil_interpolation = pil_modes_mapping[interpolation]
        return F_pil.perspective(img, coeffs, interpolation=pil_interpolation, fill=fill)
668

669
    return F_t.perspective(img, coeffs, interpolation=interpolation.value, fill=fill)
670
671


672
def vflip(img: Tensor) -> Tensor:
673
    """Vertically flip the given image.
674
675

    Args:
vfdev's avatar
vfdev committed
676
        img (PIL Image or Tensor): Image to be flipped. If img
677
            is a Tensor, it is expected to be in [..., H, W] format,
678
            where ... means it can have an arbitrary number of leading
679
            dimensions.
680
681

    Returns:
682
        PIL Image or Tensor:  Vertically flipped image.
683
    """
684
685
    if not isinstance(img, torch.Tensor):
        return F_pil.vflip(img)
686

687
    return F_t.vflip(img)
688
689


vfdev's avatar
vfdev committed
690
691
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.
692
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
693
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
694
695
696
697
698
699

    .. 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
700
701
702
        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
703
            made. If provided a sequence of length 1, it will be interpreted as (size[0], size[0]).
704

705
    Returns:
706
       tuple: tuple (tl, tr, bl, br, center)
707
       Corresponding top left, top right, bottom left, bottom right and center crop.
708
709
710
    """
    if isinstance(size, numbers.Number):
        size = (int(size), int(size))
vfdev's avatar
vfdev committed
711
712
    elif isinstance(size, (tuple, list)) and len(size) == 1:
        size = (size[0], size[0])
713

vfdev's avatar
vfdev committed
714
715
716
    if len(size) != 2:
        raise ValueError("Please provide only two dimensions (h, w) for size.")

717
    image_width, image_height = get_image_size(img)
718
719
720
721
722
    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
723
724
725
726
727
728
729
730
    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
731
732


vfdev's avatar
vfdev committed
733
734
735
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
736
    flipped version of these (horizontal flipping is used by default).
737
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
738
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions
739
740
741
742
743

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

744
    Args:
vfdev's avatar
vfdev committed
745
        img (PIL Image or Tensor): Image to be cropped.
746
        size (sequence or int): Desired output size of the crop. If size is an
747
            int instead of sequence like (h, w), a square crop (size, size) is
748
            made. If provided a sequence of length 1, it will be interpreted as (size[0], size[0]).
749
        vertical_flip (bool): Use vertical flipping instead of horizontal
750
751

    Returns:
752
        tuple: tuple (tl, tr, bl, br, center, tl_flip, tr_flip, bl_flip, br_flip, center_flip)
753
754
        Corresponding top left, top right, bottom left, bottom right and
        center crop and same for the flipped image.
755
756
757
    """
    if isinstance(size, numbers.Number):
        size = (int(size), int(size))
vfdev's avatar
vfdev committed
758
759
760
761
762
    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.")
763
764
765
766
767
768
769
770
771
772
773
774

    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


775
def adjust_brightness(img: Tensor, brightness_factor: float) -> Tensor:
776
    """Adjust brightness 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 [..., 1 or 3, H, W] format,
            where ... means it can have an arbitrary number of leading dimensions.
782
783
784
785
786
        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
787
        PIL Image or Tensor: Brightness adjusted image.
788
    """
789
790
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_brightness(img, brightness_factor)
791

792
    return F_t.adjust_brightness(img, brightness_factor)
793
794


795
def adjust_contrast(img: Tensor, contrast_factor: float) -> Tensor:
796
    """Adjust contrast of an image.
797
798

    Args:
vfdev's avatar
vfdev committed
799
        img (PIL Image or Tensor): Image to be adjusted.
800
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
801
            where ... means it can have an arbitrary number of leading dimensions.
802
803
804
805
806
        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
807
        PIL Image or Tensor: Contrast adjusted image.
808
    """
809
810
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_contrast(img, contrast_factor)
811

812
    return F_t.adjust_contrast(img, contrast_factor)
813
814


815
def adjust_saturation(img: Tensor, saturation_factor: float) -> Tensor:
816
817
818
    """Adjust color saturation of an image.

    Args:
vfdev's avatar
vfdev committed
819
        img (PIL Image or Tensor): Image to be adjusted.
820
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
821
            where ... means it can have an arbitrary number of leading dimensions.
822
823
824
825
826
        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
827
        PIL Image or Tensor: Saturation adjusted image.
828
    """
829
830
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_saturation(img, saturation_factor)
831

832
    return F_t.adjust_saturation(img, saturation_factor)
833
834


835
def adjust_hue(img: Tensor, hue_factor: float) -> Tensor:
836
837
838
839
840
841
842
843
844
    """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]`.

845
846
847
    See `Hue`_ for more details.

    .. _Hue: https://en.wikipedia.org/wiki/Hue
848
849

    Args:
850
        img (PIL Image or Tensor): Image to be adjusted.
851
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
852
            where ... means it can have an arbitrary number of leading dimensions.
853
            If img is PIL Image mode "1", "I", "F" and modes with transparency (alpha channel) are not supported.
854
855
856
857
858
859
860
        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:
861
        PIL Image or Tensor: Hue adjusted image.
862
    """
863
864
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_hue(img, hue_factor)
865

866
    return F_t.adjust_hue(img, hue_factor)
867
868


869
def adjust_gamma(img: Tensor, gamma: float, gain: float = 1) -> Tensor:
870
    r"""Perform gamma correction on an image.
871
872
873
874

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

875
876
877
878
    .. math::
        I_{\text{out}} = 255 \times \text{gain} \times \left(\frac{I_{\text{in}}}{255}\right)^{\gamma}

    See `Gamma Correction`_ for more details.
879

880
    .. _Gamma Correction: https://en.wikipedia.org/wiki/Gamma_correction
881
882

    Args:
883
        img (PIL Image or Tensor): PIL Image to be adjusted.
884
885
886
            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.
887
888
889
        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.
890
        gain (float): The constant multiplier.
891
892
    Returns:
        PIL Image or Tensor: Gamma correction adjusted image.
893
    """
894
895
    if not isinstance(img, torch.Tensor):
        return F_pil.adjust_gamma(img, gamma, gain)
896

897
    return F_t.adjust_gamma(img, gamma, gain)
898
899


vfdev's avatar
vfdev committed
900
def _get_inverse_affine_matrix(
901
    center: List[float], angle: float, translate: List[float], scale: float, shear: List[float]
vfdev's avatar
vfdev committed
902
) -> List[float]:
903
904
905
906
907
908
909
    # 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
910
911
912
913
914
915
916
917
918
919
    #       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]
    #
920
921
    # Thus, the inverse is M^-1 = C * RSS^-1 * C^-1 * T^-1

922
    rot = math.radians(angle)
923
924
    sx = math.radians(shear[0])
    sy = math.radians(shear[1])
925
926
927
928
929

    cx, cy = center
    tx, ty = translate

    # RSS without scaling
vfdev's avatar
vfdev committed
930
931
932
933
    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)
934
935

    # Inverted rotation matrix with scale and shear
936
    # det([[a, b], [c, d]]) == 1, since det(rotation) = 1 and det(shear) = 1
vfdev's avatar
vfdev committed
937
938
    matrix = [d, -b, 0.0, -c, a, 0.0]
    matrix = [x / scale for x in matrix]
939
940

    # Apply inverse of translation and of center translation: RSS^-1 * C^-1 * T^-1
vfdev's avatar
vfdev committed
941
942
    matrix[2] += matrix[0] * (-cx - tx) + matrix[1] * (-cy - ty)
    matrix[5] += matrix[3] * (-cx - tx) + matrix[4] * (-cy - ty)
943
944

    # Apply center translation: C * RSS^-1 * C^-1 * T^-1
vfdev's avatar
vfdev committed
945
946
    matrix[2] += cx
    matrix[5] += cy
947

vfdev's avatar
vfdev committed
948
    return matrix
949

vfdev's avatar
vfdev committed
950

vfdev's avatar
vfdev committed
951
def rotate(
952
953
954
955
956
957
958
    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
959
960
) -> Tensor:
    """Rotate the image by angle.
961
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
962
963
964
965
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.

    Args:
        img (PIL Image or Tensor): image to be rotated.
966
        angle (number): rotation angle value in degrees, counter-clockwise.
967
968
969
        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.
970
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
vfdev's avatar
vfdev committed
971
972
973
974
        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.
975
        center (sequence, optional): Optional center of rotation. Origin is the upper left corner.
vfdev's avatar
vfdev committed
976
            Default is the center of the image.
977
978
        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.
979
980
981
982

            .. note::
                In torchscript mode single int/float value is not supported, please use a sequence
                of length 1: ``[value, ]``.
vfdev's avatar
vfdev committed
983
984
985
986
987
988
989

    Returns:
        PIL Image or Tensor: Rotated image.

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

    """
990
991
992
993
994
995
996
997
998
    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(
999
1000
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
1001
1002
1003
        )
        interpolation = _interpolation_modes_from_int(interpolation)

vfdev's avatar
vfdev committed
1004
1005
1006
1007
1008
1009
    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")

1010
1011
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
1012

vfdev's avatar
vfdev committed
1013
    if not isinstance(img, torch.Tensor):
1014
1015
        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
1016
1017
1018

    center_f = [0.0, 0.0]
    if center is not None:
1019
        img_size = get_image_size(img)
1020
1021
1022
        # 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
1023
1024
1025
    # 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])
1026
    return F_t.rotate(img, matrix=matrix, interpolation=interpolation.value, expand=expand, fill=fill)
vfdev's avatar
vfdev committed
1027
1028


vfdev's avatar
vfdev committed
1029
def affine(
1030
1031
1032
1033
1034
1035
1036
1037
1038
    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,
vfdev's avatar
vfdev committed
1039
1040
) -> Tensor:
    """Apply affine transformation on the image keeping image center invariant.
1041
    If the image is torch Tensor, it is expected
vfdev's avatar
vfdev committed
1042
    to have [..., H, W] shape, where ... means an arbitrary number of leading dimensions.
1043
1044

    Args:
vfdev's avatar
vfdev committed
1045
        img (PIL Image or Tensor): image to transform.
1046
1047
        angle (number): rotation angle in degrees between -180 and 180, clockwise direction.
        translate (sequence of integers): horizontal and vertical translations (post-rotation translation)
1048
        scale (float): overall scale
1049
1050
        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
1051
            the second value corresponds to a shear parallel to the y axis.
1052
1053
1054
        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.
1055
            For backward compatibility integer values (e.g. ``PIL.Image.NEAREST``) are still acceptable.
1056
1057
        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.
1058
1059
1060
1061

            .. note::
                In torchscript mode single int/float value is not supported, please use a sequence
                of length 1: ``[value, ]``.
1062
        fillcolor (sequence, int, float): deprecated argument and will be removed since v0.10.0.
1063
            Please use the ``fill`` parameter instead.
1064
        resample (int, optional): deprecated argument and will be removed since v0.10.0.
1065
            Please use the ``interpolation`` parameter instead.
vfdev's avatar
vfdev committed
1066
1067
1068

    Returns:
        PIL Image or Tensor: Transformed image.
1069
    """
1070
1071
1072
1073
1074
1075
1076
1077
1078
    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(
1079
1080
            "Argument interpolation should be of type InterpolationMode instead of int. "
            "Please, use InterpolationMode enum."
1081
1082
1083
1084
        )
        interpolation = _interpolation_modes_from_int(interpolation)

    if fillcolor is not None:
1085
        warnings.warn("Argument fillcolor is deprecated and will be removed since v0.10.0. Please, use fill instead")
1086
1087
        fill = fillcolor

vfdev's avatar
vfdev committed
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
1099
1100
1101
1102
    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")

1103
1104
    if not isinstance(interpolation, InterpolationMode):
        raise TypeError("Argument interpolation should be a InterpolationMode")
1105

vfdev's avatar
vfdev committed
1106
1107
1108
1109
1110
1111
1112
1113
1114
1115
1116
1117
1118
1119
1120
1121
    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:
1122
        raise ValueError(f"Shear should be a sequence containing two values. Got {shear}")
vfdev's avatar
vfdev committed
1123

1124
    img_size = get_image_size(img)
vfdev's avatar
vfdev committed
1125
1126
1127
1128
1129
1130
    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)
1131
1132
        pil_interpolation = pil_modes_mapping[interpolation]
        return F_pil.affine(img, matrix=matrix, interpolation=pil_interpolation, fill=fill)
1133

1134
1135
    translate_f = [1.0 * t for t in translate]
    matrix = _get_inverse_affine_matrix([0.0, 0.0], angle, translate_f, scale, shear)
1136
    return F_t.affine(img, matrix=matrix, interpolation=interpolation.value, fill=fill)
1137
1138


1139
@torch.jit.unused
1140
def to_grayscale(img, num_output_channels=1):
1141
    """Convert PIL image of any mode (RGB, HSV, LAB, etc) to grayscale version of image.
1142
    This transform does not support torch Tensor.
1143
1144

    Args:
1145
        img (PIL Image): PIL Image to be converted to grayscale.
1146
        num_output_channels (int): number of channels of the output image. Value can be 1 or 3. Default is 1.
1147
1148

    Returns:
1149
1150
        PIL Image: Grayscale version of the image.

1151
1152
        - if num_output_channels = 1 : returned image is single channel
        - if num_output_channels = 3 : returned image is 3 channel with r = g = b
1153
    """
1154
1155
    if isinstance(img, Image.Image):
        return F_pil.to_grayscale(img, num_output_channels)
1156

1157
1158
1159
1160
1161
    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.
1162
1163
    If the image is torch Tensor, it is expected
    to have [..., 3, H, W] shape, where ... means an arbitrary number of leading dimensions
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
1174
1175

    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.

1176
1177
        - if num_output_channels = 1 : returned image is single channel
        - if num_output_channels = 3 : returned image is 3 channel with r = g = b
1178
1179
1180
1181
1182
    """
    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)
1183
1184


1185
def erase(img: Tensor, i: int, j: int, h: int, w: int, v: Tensor, inplace: bool = False) -> Tensor:
1186
    """Erase the input Tensor Image with given value.
1187
    This transform does not support PIL Image.
1188
1189
1190
1191
1192
1193
1194
1195

    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
1196
        inplace(bool, optional): For in-place operations. By default is set False.
1197
1198
1199
1200
1201

    Returns:
        Tensor Image: Erased image.
    """
    if not isinstance(img, torch.Tensor):
1202
        raise TypeError(f"img should be Tensor Image. Got {type(img)}")
1203

1204
1205
1206
    if not inplace:
        img = img.clone()

1207
    img[..., i : i + h, j : j + w] = v
1208
    return img
1209
1210
1211


def gaussian_blur(img: Tensor, kernel_size: List[int], sigma: Optional[List[float]] = None) -> Tensor:
1212
1213
1214
    """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.
1215
1216
1217
1218
1219

    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.
1220
1221
1222
1223

            .. note::
                In torchscript mode kernel_size as single int is not supported, use a sequence of
                length 1: ``[ksize, ]``.
1224
1225
1226
1227
        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``.
1228
1229
1230
1231
1232
            Default, None.

            .. note::
                In torchscript mode sigma as single float is
                not supported, use a sequence of length 1: ``[sigma, ]``.
1233
1234
1235
1236
1237

    Returns:
        PIL Image or Tensor: Gaussian Blurred version of the image.
    """
    if not isinstance(kernel_size, (int, list, tuple)):
1238
        raise TypeError(f"kernel_size should be int or a sequence of integers. Got {type(kernel_size)}")
1239
1240
1241
    if isinstance(kernel_size, int):
        kernel_size = [kernel_size, kernel_size]
    if len(kernel_size) != 2:
1242
        raise ValueError(f"If kernel_size is a sequence its length should be 2. Got {len(kernel_size)}")
1243
1244
    for ksize in kernel_size:
        if ksize % 2 == 0 or ksize < 0:
1245
            raise ValueError(f"kernel_size should have odd and positive integers. Got {kernel_size}")
1246
1247
1248
1249
1250

    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)):
1251
        raise TypeError(f"sigma should be either float or sequence of floats. Got {type(sigma)}")
1252
1253
1254
1255
1256
    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:
1257
        raise ValueError(f"If sigma is a sequence, its length should be 2. Got {len(sigma)}")
1258
    for s in sigma:
1259
        if s <= 0.0:
1260
            raise ValueError(f"sigma should have positive values. Got {sigma}")
1261
1262
1263
1264

    t_img = img
    if not isinstance(img, torch.Tensor):
        if not F_pil._is_pil_image(img):
1265
            raise TypeError(f"img should be PIL Image or Tensor. Got {type(img)}")
1266
1267
1268
1269
1270
1271
1272
1273

        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
1274
1275
1276


def invert(img: Tensor) -> Tensor:
1277
    """Invert the colors of an RGB/grayscale image.
1278
1279
1280

    Args:
        img (PIL Image or Tensor): Image to have its colors inverted.
1281
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1282
1283
            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".
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294

    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:
1295
    """Posterize an image by reducing the number of bits for each color channel.
1296
1297
1298

    Args:
        img (PIL Image or Tensor): Image to have its colors posterized.
1299
            If img is torch Tensor, it should be of type torch.uint8 and
1300
1301
1302
            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".
1303
1304
1305
1306
1307
        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):
1308
        raise ValueError(f"The number if bits should be between 0 and 8. Got {bits}")
1309
1310
1311
1312
1313
1314
1315
1316

    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:
1317
    """Solarize an RGB/grayscale image by inverting all pixel values above a threshold.
1318
1319
1320

    Args:
        img (PIL Image or Tensor): Image to have its colors inverted.
1321
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1322
1323
            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".
1324
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
        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:
1335
    """Adjust the sharpness of an image.
1336
1337
1338

    Args:
        img (PIL Image or Tensor): Image to be adjusted.
1339
1340
            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.
1341
1342
1343
1344
1345
1346
1347
1348
1349
1350
1351
1352
1353
1354
        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:
1355
    """Maximize contrast of an image by remapping its
1356
1357
1358
1359
1360
    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.
1361
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1362
1363
            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".
1364
1365
1366
1367
1368
1369
1370
1371
1372
1373
1374

    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:
1375
    """Equalize the histogram of an image by applying
1376
1377
1378
1379
1380
    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.
1381
            If img is torch Tensor, it is expected to be in [..., 1 or 3, H, W] format,
1382
            where ... means it can have an arbitrary number of leading dimensions.
1383
            The tensor dtype must be ``torch.uint8`` and values are expected to be in ``[0, 255]``.
1384
            If img is PIL Image, it is expected to be in mode "P", "L" or "RGB".
1385
1386
1387
1388
1389
1390
1391
1392

    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)