File size: 19,534 Bytes
9c6594c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
# Copyright The Lightning team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
#     http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from collections.abc import Sequence
from typing import Any, List, Optional, Union

import torch
from torch import Tensor
from typing_extensions import Literal

from torchmetrics.functional.image.ssim import _multiscale_ssim_update, _ssim_check_inputs, _ssim_update
from torchmetrics.metric import Metric
from torchmetrics.utilities.data import dim_zero_cat
from torchmetrics.utilities.imports import _MATPLOTLIB_AVAILABLE
from torchmetrics.utilities.plot import _AX_TYPE, _PLOT_OUT_TYPE

if not _MATPLOTLIB_AVAILABLE:
    __doctest_skip__ = ["StructuralSimilarityIndexMeasure.plot", "MultiScaleStructuralSimilarityIndexMeasure.plot"]


class StructuralSimilarityIndexMeasure(Metric):
    """Compute Structural Similarity Index Measure (SSIM_).

    As input to ``forward`` and ``update`` the metric accepts the following input

    - ``preds`` (:class:`~torch.Tensor`): Predictions from model
    - ``target`` (:class:`~torch.Tensor`): Ground truth values

    As output of `forward` and `compute` the metric returns the following output

    - ``ssim`` (:class:`~torch.Tensor`): if ``reduction!='none'`` returns float scalar tensor with average SSIM value
      over sample else returns tensor of shape ``(N,)`` with SSIM values per sample

    Args:
        preds: estimated image
        target: ground truth image
        gaussian_kernel: If ``True`` (default), a gaussian kernel is used, if ``False`` a uniform kernel is used
        sigma: Standard deviation of the gaussian kernel, anisotropic kernels are possible.
            Ignored if a uniform kernel is used
        kernel_size: the size of the uniform kernel, anisotropic kernels are possible.
            Ignored if a Gaussian kernel is used
        reduction: a method to reduce metric score over individual batch scores

            - ``'elementwise_mean'``: takes the mean
            - ``'sum'``: takes the sum
            - ``'none'`` or ``None``: no reduction will be applied

        data_range:
            the range of the data. If None, it is determined from the data (max - min). If a tuple is provided then
            the range is calculated as the difference and input is clamped between the values.
        k1: Parameter of SSIM.
        k2: Parameter of SSIM.
        return_full_image: If true, the full ``ssim`` image is returned as a second argument.
            Mutually exclusive with ``return_contrast_sensitivity``
        return_contrast_sensitivity: If true, the constant term is returned as a second argument.
            The luminance term can be obtained with luminance=ssim/contrast
            Mutually exclusive with ``return_full_image``
        kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.

    Example:
        >>> import torch
        >>> from torchmetrics.image import StructuralSimilarityIndexMeasure
        >>> preds = torch.rand([3, 3, 256, 256])
        >>> target = preds * 0.75
        >>> ssim = StructuralSimilarityIndexMeasure(data_range=1.0)
        >>> ssim(preds, target)
        tensor(0.9219)

    """

    higher_is_better: bool = True
    is_differentiable: bool = True
    full_state_update: bool = False
    plot_lower_bound: float = 0.0
    plot_upper_bound: float = 1.0

    preds: List[Tensor]
    target: List[Tensor]

    def __init__(
        self,
        gaussian_kernel: bool = True,
        sigma: Union[float, Sequence[float]] = 1.5,
        kernel_size: Union[int, Sequence[int]] = 11,
        reduction: Literal["elementwise_mean", "sum", "none", None] = "elementwise_mean",
        data_range: Optional[Union[float, tuple[float, float]]] = None,
        k1: float = 0.01,
        k2: float = 0.03,
        return_full_image: bool = False,
        return_contrast_sensitivity: bool = False,
        **kwargs: Any,
    ) -> None:
        super().__init__(**kwargs)

        valid_reduction = ("elementwise_mean", "sum", "none", None)
        if reduction not in valid_reduction:
            raise ValueError(f"Argument `reduction` must be one of {valid_reduction}, but got {reduction}")

        if reduction in ("elementwise_mean", "sum"):
            self.add_state("similarity", default=torch.tensor(0.0), dist_reduce_fx="sum")
        else:
            self.add_state("similarity", default=[], dist_reduce_fx="cat")

        self.add_state("total", default=torch.tensor(0.0), dist_reduce_fx="sum")

        if return_contrast_sensitivity or return_full_image:
            self.add_state("image_return", default=[], dist_reduce_fx="cat")

        self.gaussian_kernel = gaussian_kernel
        self.sigma = sigma
        self.kernel_size = kernel_size
        self.reduction = reduction
        self.data_range = data_range
        self.k1 = k1
        self.k2 = k2
        self.return_full_image = return_full_image
        self.return_contrast_sensitivity = return_contrast_sensitivity

    def update(self, preds: Tensor, target: Tensor) -> None:
        """Update state with predictions and targets."""
        preds, target = _ssim_check_inputs(preds, target)
        similarity_pack = _ssim_update(
            preds,
            target,
            self.gaussian_kernel,
            self.sigma,
            self.kernel_size,
            self.data_range,
            self.k1,
            self.k2,
            self.return_full_image,
            self.return_contrast_sensitivity,
        )

        if isinstance(similarity_pack, tuple):
            similarity, image = similarity_pack
        else:
            similarity = similarity_pack

        if self.return_contrast_sensitivity or self.return_full_image:
            if not isinstance(self.image_return, list):
                raise TypeError("Expected `self.image_return` to be a list when returning images.")
            self.image_return.append(image)

        if self.reduction in ("elementwise_mean", "sum"):
            if not isinstance(self.similarity, torch.Tensor):  # Ensure it's a Tensor
                raise TypeError("Expected `self.similarity` to be a Tensor for reductions.")
            self.similarity += similarity.sum()
            if not isinstance(self.total, torch.Tensor):
                raise TypeError("Expected `self.total` to be a Tensor.")
            self.total += preds.shape[0]
        else:
            if not isinstance(self.similarity, list):
                raise TypeError("Expected `self.similarity` to be a list when reduction='none'.")
            self.similarity.append(similarity)

    def compute(self) -> Union[Tensor, tuple[Tensor, Tensor]]:
        """Compute SSIM over state."""
        if self.reduction == "elementwise_mean":
            if isinstance(self.similarity, Tensor) and isinstance(self.total, Tensor):
                similarity = self.similarity / self.total
            else:
                raise TypeError(
                    "Expected `self.similarity`and `self.total` to be of type Tensor for elementwise_mean reduction."
                )
        elif self.reduction == "sum":
            if not isinstance(self.similarity, Tensor):
                raise TypeError("Expected `self.similarity` to be a Tensor for sum reduction.")
            similarity = self.similarity
        else:
            if isinstance(self.similarity, list):
                similarity = dim_zero_cat(self.similarity)  # Concatenate list of Tensors
            else:
                raise TypeError("Expected `self.similarity` to be a list for reduction='none'.")

        if self.return_contrast_sensitivity or self.return_full_image:
            if isinstance(self.image_return, list):
                image_return = dim_zero_cat(self.image_return)  # Concatenate list of Tensors
            else:
                raise TypeError("Expected `self.image_return` to be a list when returning images.")
            return similarity, image_return

        return similarity

    def plot(
        self, val: Optional[Union[Tensor, Sequence[Tensor]]] = None, ax: Optional[_AX_TYPE] = None
    ) -> _PLOT_OUT_TYPE:
        """Plot a single or multiple values from the metric.

        Args:
            val: Either a single result from calling `metric.forward` or `metric.compute` or a list of these results.
                If no value is provided, will automatically call `metric.compute` and plot that result.
            ax: An matplotlib axis object. If provided will add plot to that axis

        Returns:
            Figure and Axes object

        Raises:
            ModuleNotFoundError:
                If `matplotlib` is not installed

        .. plot::
            :scale: 75

            >>> # Example plotting a single value
            >>> import torch
            >>> from torchmetrics.image import StructuralSimilarityIndexMeasure
            >>> preds = torch.rand([3, 3, 256, 256])
            >>> target = preds * 0.75
            >>> metric = StructuralSimilarityIndexMeasure(data_range=1.0)
            >>> metric.update(preds, target)
            >>> fig_, ax_ = metric.plot()

        .. plot::
            :scale: 75

            >>> # Example plotting multiple values
            >>> import torch
            >>> from torchmetrics.image import StructuralSimilarityIndexMeasure
            >>> preds = torch.rand([3, 3, 256, 256])
            >>> target = preds * 0.75
            >>> metric = StructuralSimilarityIndexMeasure(data_range=1.0)
            >>> values = [ ]
            >>> for _ in range(10):
            ...     values.append(metric(preds, target))
            >>> fig_, ax_ = metric.plot(values)

        """
        return self._plot(val, ax)


class MultiScaleStructuralSimilarityIndexMeasure(Metric):
    """Compute `MultiScaleSSIM`_, Multi-scale Structural Similarity Index Measure.

    This metric is is a generalization of Structural Similarity Index Measure by incorporating image details at
    different resolution scores.

    As input to ``forward`` and ``update`` the metric accepts the following input

    - ``preds`` (:class:`~torch.Tensor`): Predictions from model
    - ``target`` (:class:`~torch.Tensor`): Ground truth values

    As output of `forward` and `compute` the metric returns the following output

    - ``msssim`` (:class:`~torch.Tensor`): if ``reduction!='none'`` returns float scalar tensor with average MSSSIM
      value over sample else returns tensor of shape ``(N,)`` with SSIM values per sample

    Args:
        gaussian_kernel: If ``True`` (default), a gaussian kernel is used, if false a uniform kernel is used
        kernel_size: size of the gaussian kernel
        sigma: Standard deviation of the gaussian kernel
        reduction: a method to reduce metric score over labels.

            - ``'elementwise_mean'``: takes the mean
            - ``'sum'``: takes the sum
            - ``'none'`` or ``None``: no reduction will be applied

        data_range:
            the range of the data. If None, it is determined from the data (max - min). If a tuple is provided then
            the range is calculated as the difference and input is clamped between the values.
            The ``data_range`` must be given when ``dim`` is not None.
        k1: Parameter of structural similarity index measure.
        k2: Parameter of structural similarity index measure.
        betas: Exponent parameters for individual similarities and contrastive sensitivities returned by different image
            resolutions.
        normalize: When MultiScaleStructuralSimilarityIndexMeasure loss is used for training, it is desirable to use
            normalizes to improve the training stability. This `normalize` argument is out of scope of the original
            implementation [1], and it is adapted from https://github.com/jorge-pessoa/pytorch-msssim instead.
        kwargs: Additional keyword arguments, see :ref:`Metric kwargs` for more info.

    Return:
        Tensor with Multi-Scale SSIM score

    Raises:
        ValueError:
            If ``kernel_size`` is not an int or a Sequence of ints with size 2 or 3.
        ValueError:
            If ``betas`` is not a tuple of floats with length 2.
        ValueError:
            If ``normalize`` is neither `None`, `ReLU` nor `simple`.

    Example:
        >>> from torch import rand
        >>> from torchmetrics.image import MultiScaleStructuralSimilarityIndexMeasure
        >>> preds = torch.rand([3, 3, 256, 256])
        >>> target = preds * 0.75
        >>> ms_ssim = MultiScaleStructuralSimilarityIndexMeasure(data_range=1.0)
        >>> ms_ssim(preds, target)
        tensor(0.9628)

    """

    higher_is_better: bool = True
    is_differentiable: bool = True
    full_state_update: bool = False
    plot_lower_bound: float = 0.0
    plot_upper_bound: float = 1.0

    preds: List[Tensor]
    target: List[Tensor]

    def __init__(
        self,
        gaussian_kernel: bool = True,
        kernel_size: Union[int, Sequence[int]] = 11,
        sigma: Union[float, Sequence[float]] = 1.5,
        reduction: Literal["elementwise_mean", "sum", "none", None] = "elementwise_mean",
        data_range: Optional[Union[float, tuple[float, float]]] = None,
        k1: float = 0.01,
        k2: float = 0.03,
        betas: tuple[float, ...] = (0.0448, 0.2856, 0.3001, 0.2363, 0.1333),
        normalize: Literal["relu", "simple", None] = "relu",
        **kwargs: Any,
    ) -> None:
        super().__init__(**kwargs)

        valid_reduction = ("elementwise_mean", "sum", "none", None)
        if reduction not in valid_reduction:
            raise ValueError(f"Argument `reduction` must be one of {valid_reduction}, but got {reduction}")

        if reduction in ("elementwise_mean", "sum"):
            self.add_state("similarity", default=torch.tensor(0.0), dist_reduce_fx="sum")
        else:
            self.add_state("similarity", default=[], dist_reduce_fx="cat")

        self.add_state("total", default=torch.tensor(0.0), dist_reduce_fx="sum")

        if not (isinstance(kernel_size, (Sequence, int))):
            raise ValueError(
                f"Argument `kernel_size` expected to be an sequence or an int, or a single int. Got {kernel_size}"
            )
        if isinstance(kernel_size, Sequence) and (
            len(kernel_size) not in (2, 3) or not all(isinstance(ks, int) for ks in kernel_size)
        ):
            raise ValueError(
                "Argument `kernel_size` expected to be an sequence of size 2 or 3 where each element is an int, "
                f"or a single int. Got {kernel_size}"
            )

        self.gaussian_kernel = gaussian_kernel
        self.sigma = sigma
        self.kernel_size = kernel_size
        self.reduction = reduction
        self.data_range = data_range
        self.k1 = k1
        self.k2 = k2
        if not isinstance(betas, tuple):
            raise ValueError("Argument `betas` is expected to be of a type tuple.")
        if isinstance(betas, tuple) and not all(isinstance(beta, float) for beta in betas):
            raise ValueError("Argument `betas` is expected to be a tuple of floats.")
        self.betas = betas
        if normalize and normalize not in ("relu", "simple"):
            raise ValueError("Argument `normalize` to be expected either `None` or one of 'relu' or 'simple'")
        self.normalize = normalize

    def update(self, preds: Tensor, target: Tensor) -> None:
        """Update state with predictions and targets."""
        preds, target = _ssim_check_inputs(preds, target)
        similarity = _multiscale_ssim_update(
            preds,
            target,
            self.gaussian_kernel,
            self.sigma,
            self.kernel_size,
            self.data_range,
            self.k1,
            self.k2,
            self.betas,
            self.normalize,
        )

        if self.reduction in ("none", None):
            if not isinstance(self.similarity, list):
                raise TypeError("Expected `self.similarity` to be a list for reduction='none'.")
            self.similarity.append(similarity)
        else:
            if not isinstance(self.similarity, Tensor):
                raise TypeError("Expected `self.similarity` to be a Tensor for elementwise_mean or sum reduction.")
            self.similarity += similarity.sum()

        if not isinstance(self.total, Tensor):
            raise TypeError("Expected `self.total` to be a Tensor.")
        self.total += torch.tensor(preds.shape[0], dtype=self.total.dtype, device=self.total.device)

    def compute(self) -> Tensor:
        """Compute MS-SSIM over state."""
        if self.reduction in ("none", None):
            if isinstance(self.similarity, list):
                return dim_zero_cat(self.similarity)
            raise TypeError("Expected `self.similarity` to be a list for reduction='none'.")
        if self.reduction == "sum":
            if isinstance(self.similarity, Tensor):
                return self.similarity
            raise TypeError("Expected `self.similarity` to be a Tensor for sum reduction.")
        if isinstance(self.similarity, Tensor) and isinstance(self.total, Tensor):
            return self.similarity / self.total
        raise TypeError("Expected `self.similarity` and `self.total` to be Tensors for elementwise_mean reduction.")

    def plot(
        self, val: Optional[Union[Tensor, Sequence[Tensor]]] = None, ax: Optional[_AX_TYPE] = None
    ) -> _PLOT_OUT_TYPE:
        """Plot a single or multiple values from the metric.

        Args:
            val: Either a single result from calling `metric.forward` or `metric.compute` or a list of these results.
                If no value is provided, will automatically call `metric.compute` and plot that result.
            ax: An matplotlib axis object. If provided will add plot to that axis

        Returns:
            Figure and Axes object

        Raises:
            ModuleNotFoundError:
                If `matplotlib` is not installed

        .. plot::
            :scale: 75

            >>> # Example plotting a single value
            >>> from torch import rand
            >>> from torchmetrics.image import MultiScaleStructuralSimilarityIndexMeasure
            >>> preds = rand([3, 3, 256, 256])
            >>> target = preds * 0.75
            >>> metric = MultiScaleStructuralSimilarityIndexMeasure(data_range=1.0)
            >>> metric.update(preds, target)
            >>> fig_, ax_ = metric.plot()

        .. plot::
            :scale: 75

            >>> # Example plotting multiple values
            >>> from torch import rand
            >>> from torchmetrics.image import MultiScaleStructuralSimilarityIndexMeasure
            >>> preds = rand([3, 3, 256, 256])
            >>> target = preds * 0.75
            >>> metric = MultiScaleStructuralSimilarityIndexMeasure(data_range=1.0)
            >>> values = [ ]
            >>> for _ in range(10):
            ...     values.append(metric(preds, target))
            >>> fig_, ax_ = metric.plot(values)

        """
        return self._plot(val, ax)