test_transforms.py 80.8 KB
Newer Older
1
import os
2
3
import torch
import torchvision.transforms as transforms
4
import torchvision.transforms.functional as F
5
import torchvision.transforms.functional_tensor as F_t
6
from torch._utils_internal import get_file_path_2
7
from numpy.testing import assert_array_almost_equal
8
import unittest
9
import math
10
import random
11
import numpy as np
12
13
14
15
16
17
from PIL import Image
try:
    import accimage
except ImportError:
    accimage = None

18
19
20
21
22
try:
    from scipy import stats
except ImportError:
    stats = None

23
from common_utils import cycle_over, int_dtypes, float_dtypes
24
25


26
GRACE_HOPPER = get_file_path_2(
27
    os.path.dirname(os.path.abspath(__file__)), 'assets', 'encode_jpeg', 'grace_hopper_517x606.jpg')
28
29


30
class Tester(unittest.TestCase):
31

32
33
34
35
    def test_crop(self):
        height = random.randint(10, 32) * 2
        width = random.randint(10, 32) * 2
        oheight = random.randint(5, (height - 2) / 2) * 2
36
37
        owidth = random.randint(5, (width - 2) / 2) * 2

38
        img = torch.ones(3, height, width)
39
40
41
        oh1 = (height - oheight) // 2
        ow1 = (width - owidth) // 2
        imgnarrow = img[:, oh1:oh1 + oheight, ow1:ow1 + owidth]
42
43
44
45
46
47
        imgnarrow.fill_(0)
        result = transforms.Compose([
            transforms.ToPILImage(),
            transforms.CenterCrop((oheight, owidth)),
            transforms.ToTensor(),
        ])(img)
48
49
        self.assertEqual(result.sum(), 0,
                         "height: {} width: {} oheight: {} owdith: {}".format(height, width, oheight, owidth))
50
51
52
53
54
55
56
57
        oheight += 1
        owidth += 1
        result = transforms.Compose([
            transforms.ToPILImage(),
            transforms.CenterCrop((oheight, owidth)),
            transforms.ToTensor(),
        ])(img)
        sum1 = result.sum()
58
59
        self.assertGreater(sum1, 1,
                           "height: {} width: {} oheight: {} owdith: {}".format(height, width, oheight, owidth))
60
        oheight += 1
61
        owidth += 1
62
63
64
65
66
67
        result = transforms.Compose([
            transforms.ToPILImage(),
            transforms.CenterCrop((oheight, owidth)),
            transforms.ToTensor(),
        ])(img)
        sum2 = result.sum()
68
69
70
71
        self.assertGreater(sum2, 0,
                           "height: {} width: {} oheight: {} owdith: {}".format(height, width, oheight, owidth))
        self.assertGreater(sum2, sum1,
                           "height: {} width: {} oheight: {} owdith: {}".format(height, width, oheight, owidth))
72

73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
    def test_five_crop(self):
        to_pil_image = transforms.ToPILImage()
        h = random.randint(5, 25)
        w = random.randint(5, 25)
        for single_dim in [True, False]:
            crop_h = random.randint(1, h)
            crop_w = random.randint(1, w)
            if single_dim:
                crop_h = min(crop_h, crop_w)
                crop_w = crop_h
                transform = transforms.FiveCrop(crop_h)
            else:
                transform = transforms.FiveCrop((crop_h, crop_w))

            img = torch.FloatTensor(3, h, w).uniform_()
            results = transform(to_pil_image(img))

90
            self.assertEqual(len(results), 5)
91
            for crop in results:
92
                self.assertEqual(crop.size, (crop_w, crop_h))
93
94
95
96
97
98
99
100

            to_pil_image = transforms.ToPILImage()
            tl = to_pil_image(img[:, 0:crop_h, 0:crop_w])
            tr = to_pil_image(img[:, 0:crop_h, w - crop_w:])
            bl = to_pil_image(img[:, h - crop_h:, 0:crop_w])
            br = to_pil_image(img[:, h - crop_h:, w - crop_w:])
            center = transforms.CenterCrop((crop_h, crop_w))(to_pil_image(img))
            expected_output = (tl, tr, bl, br, center)
101
            self.assertEqual(results, expected_output)
102
103
104
105
106
107
108
109
110
111
112
113

    def test_ten_crop(self):
        to_pil_image = transforms.ToPILImage()
        h = random.randint(5, 25)
        w = random.randint(5, 25)
        for should_vflip in [True, False]:
            for single_dim in [True, False]:
                crop_h = random.randint(1, h)
                crop_w = random.randint(1, w)
                if single_dim:
                    crop_h = min(crop_h, crop_w)
                    crop_w = crop_h
114
115
                    transform = transforms.TenCrop(crop_h,
                                                   vertical_flip=should_vflip)
116
117
                    five_crop = transforms.FiveCrop(crop_h)
                else:
118
119
                    transform = transforms.TenCrop((crop_h, crop_w),
                                                   vertical_flip=should_vflip)
120
121
122
123
124
                    five_crop = transforms.FiveCrop((crop_h, crop_w))

                img = to_pil_image(torch.FloatTensor(3, h, w).uniform_())
                results = transform(img)
                expected_output = five_crop(img)
125
126
127
128
129

                # Checking if FiveCrop and TenCrop can be printed as string
                transform.__repr__()
                five_crop.__repr__()

130
131
132
133
134
135
136
                if should_vflip:
                    vflipped_img = img.transpose(Image.FLIP_TOP_BOTTOM)
                    expected_output += five_crop(vflipped_img)
                else:
                    hflipped_img = img.transpose(Image.FLIP_LEFT_RIGHT)
                    expected_output += five_crop(hflipped_img)

137
138
                self.assertEqual(len(results), 10)
                self.assertEqual(results, expected_output)
139

140
141
142
143
144
145
146
147
    def test_randomresized_params(self):
        height = random.randint(24, 32) * 2
        width = random.randint(24, 32) * 2
        img = torch.ones(3, height, width)
        to_pil_image = transforms.ToPILImage()
        img = to_pil_image(img)
        size = 100
        epsilon = 0.05
148
        min_scale = 0.25
Francisco Massa's avatar
Francisco Massa committed
149
        for _ in range(10):
150
            scale_min = max(round(random.random(), 2), min_scale)
151
            scale_range = (scale_min, scale_min + round(random.random(), 2))
152
            aspect_min = max(round(random.random(), 2), epsilon)
153
154
            aspect_ratio_range = (aspect_min, aspect_min + round(random.random(), 2))
            randresizecrop = transforms.RandomResizedCrop(size, scale_range, aspect_ratio_range)
155
            i, j, h, w = randresizecrop.get_params(img, scale_range, aspect_ratio_range)
156
            aspect_ratio_obtained = w / h
157
158
159
160
161
162
163
            self.assertTrue((min(aspect_ratio_range) - epsilon <= aspect_ratio_obtained and
                             aspect_ratio_obtained <= max(aspect_ratio_range) + epsilon) or
                            aspect_ratio_obtained == 1.0)
            self.assertIsInstance(i, int)
            self.assertIsInstance(j, int)
            self.assertIsInstance(h, int)
            self.assertIsInstance(w, int)
164

165
    def test_randomperspective(self):
Francisco Massa's avatar
Francisco Massa committed
166
        for _ in range(10):
167
168
169
170
171
172
173
174
175
176
            height = random.randint(24, 32) * 2
            width = random.randint(24, 32) * 2
            img = torch.ones(3, height, width)
            to_pil_image = transforms.ToPILImage()
            img = to_pil_image(img)
            perp = transforms.RandomPerspective()
            startpoints, endpoints = perp.get_params(width, height, 0.5)
            tr_img = F.perspective(img, startpoints, endpoints)
            tr_img2 = F.to_tensor(F.perspective(tr_img, endpoints, startpoints))
            tr_img = F.to_tensor(tr_img)
177
178
179
180
            self.assertEqual(img.size[0], width)
            self.assertEqual(img.size[1], height)
            self.assertGreater(torch.nn.functional.mse_loss(tr_img, F.to_tensor(img)) + 0.3,
                               torch.nn.functional.mse_loss(tr_img2, F.to_tensor(img)))
181

182
    def test_randomperspective_fill(self):
183
184
185
186
187
188
189
190

        # assert fill being either a Sequence or a Number
        with self.assertRaises(TypeError):
            transforms.RandomPerspective(fill={})

        t = transforms.RandomPerspective(fill=None)
        self.assertTrue(t.fill == 0)

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
        height = 100
        width = 100
        img = torch.ones(3, height, width)
        to_pil_image = transforms.ToPILImage()
        img = to_pil_image(img)

        modes = ("L", "RGB", "F")
        nums_bands = [len(mode) for mode in modes]
        fill = 127

        for mode, num_bands in zip(modes, nums_bands):
            img_conv = img.convert(mode)
            perspective = transforms.RandomPerspective(p=1, fill=fill)
            tr_img = perspective(img_conv)
            pixel = tr_img.getpixel((0, 0))

            if not isinstance(pixel, tuple):
                pixel = (pixel,)
            self.assertTupleEqual(pixel, tuple([fill] * num_bands))

        for mode, num_bands in zip(modes, nums_bands):
            img_conv = img.convert(mode)
            startpoints, endpoints = transforms.RandomPerspective.get_params(width, height, 0.5)
            tr_img = F.perspective(img_conv, startpoints, endpoints, fill=fill)
            pixel = tr_img.getpixel((0, 0))
216

217
218
219
220
221
222
223
224
            if not isinstance(pixel, tuple):
                pixel = (pixel,)
            self.assertTupleEqual(pixel, tuple([fill] * num_bands))

            for wrong_num_bands in set(nums_bands) - {num_bands}:
                with self.assertRaises(ValueError):
                    F.perspective(img_conv, startpoints, endpoints, fill=tuple([fill] * wrong_num_bands))

225
    def test_resize(self):
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
        input_sizes = [
            # height, width
            # square image
            (28, 28),
            (27, 27),
            # rectangular image: h < w
            (28, 34),
            (29, 35),
            # rectangular image: h > w
            (34, 28),
            (35, 29),
        ]
        test_output_sizes_1 = [
            # single integer
            22, 27, 28, 36,
            # single integer in tuple/list
            [22, ], (27, ),
        ]
        test_output_sizes_2 = [
            # two integers
            [22, 22], [22, 28], [22, 36],
            [27, 22], [36, 22], [28, 28],
            [28, 37], [37, 27], [37, 37]
        ]

        for height, width in input_sizes:
            img = Image.new("RGB", size=(width, height), color=127)

            for osize in test_output_sizes_1:

                t = transforms.Resize(osize)
                result = t(img)

                msg = "{}, {} - {}".format(height, width, osize)
                osize = osize[0] if isinstance(osize, (list, tuple)) else osize
                # If size is an int, smaller edge of the image will be matched to this number.
                # i.e, if height > width, then image will be rescaled to (size * height / width, size).
                if height < width:
                    expected_size = (int(osize * width / height), osize)  # (w, h)
                    self.assertEqual(result.size, expected_size, msg=msg)
                elif width < height:
                    expected_size = (osize, int(osize * height / width))  # (w, h)
                    self.assertEqual(result.size, expected_size, msg=msg)
                else:
                    expected_size = (osize, osize)  # (w, h)
                    self.assertEqual(result.size, expected_size, msg=msg)
273

274
275
        for height, width in input_sizes:
            img = Image.new("RGB", size=(width, height), color=127)
276

277
278
            for osize in test_output_sizes_2:
                oheight, owidth = osize
279

280
281
                t = transforms.Resize(osize)
                result = t(img)
282

283
                self.assertEqual((owidth, oheight), result.size)
284

285
286
287
288
    def test_random_crop(self):
        height = random.randint(10, 32) * 2
        width = random.randint(10, 32) * 2
        oheight = random.randint(5, (height - 2) / 2) * 2
289
        owidth = random.randint(5, (width - 2) / 2) * 2
290
291
292
293
294
295
        img = torch.ones(3, height, width)
        result = transforms.Compose([
            transforms.ToPILImage(),
            transforms.RandomCrop((oheight, owidth)),
            transforms.ToTensor(),
        ])(img)
296
297
        self.assertEqual(result.size(1), oheight)
        self.assertEqual(result.size(2), owidth)
298

299
300
301
302
303
304
        padding = random.randint(1, 20)
        result = transforms.Compose([
            transforms.ToPILImage(),
            transforms.RandomCrop((oheight, owidth), padding=padding),
            transforms.ToTensor(),
        ])(img)
305
306
        self.assertEqual(result.size(1), oheight)
        self.assertEqual(result.size(2), owidth)
307

308
309
310
311
312
        result = transforms.Compose([
            transforms.ToPILImage(),
            transforms.RandomCrop((height, width)),
            transforms.ToTensor()
        ])(img)
313
314
315
        self.assertEqual(result.size(1), height)
        self.assertEqual(result.size(2), width)
        self.assertTrue(np.allclose(img.numpy(), result.numpy()))
316

317
318
319
320
321
        result = transforms.Compose([
            transforms.ToPILImage(),
            transforms.RandomCrop((height + 1, width + 1), pad_if_needed=True),
            transforms.ToTensor(),
        ])(img)
322
323
        self.assertEqual(result.size(1), height + 1)
        self.assertEqual(result.size(2), width + 1)
324

vfdev's avatar
vfdev committed
325
326
327
328
329
        t = transforms.RandomCrop(48)
        img = torch.ones(3, 32, 32)
        with self.assertRaisesRegex(ValueError, r"Required crop size .+ is larger then input image size .+"):
            t(img)

330
331
332
333
334
    def test_pad(self):
        height = random.randint(10, 32) * 2
        width = random.randint(10, 32) * 2
        img = torch.ones(3, height, width)
        padding = random.randint(1, 20)
335
        fill = random.randint(1, 50)
336
337
        result = transforms.Compose([
            transforms.ToPILImage(),
338
            transforms.Pad(padding, fill=fill),
339
340
            transforms.ToTensor(),
        ])(img)
341
342
        self.assertEqual(result.size(1), height + 2 * padding)
        self.assertEqual(result.size(2), width + 2 * padding)
343
344
345
346
347
348
349
350
        # check that all elements in the padded region correspond
        # to the pad value
        fill_v = fill / 255
        eps = 1e-5
        self.assertTrue((result[:, :padding, :] - fill_v).abs().max() < eps)
        self.assertTrue((result[:, :, :padding] - fill_v).abs().max() < eps)
        self.assertRaises(ValueError, transforms.Pad(padding, fill=(1, 2)),
                          transforms.ToPILImage()(img))
Soumith Chintala's avatar
Soumith Chintala committed
351

352
353
354
355
356
357
358
    def test_pad_with_tuple_of_pad_values(self):
        height = random.randint(10, 32) * 2
        width = random.randint(10, 32) * 2
        img = transforms.ToPILImage()(torch.ones(3, height, width))

        padding = tuple([random.randint(1, 20) for _ in range(2)])
        output = transforms.Pad(padding)(img)
359
        self.assertEqual(output.size, (width + padding[0] * 2, height + padding[1] * 2))
360
361
362

        padding = tuple([random.randint(1, 20) for _ in range(4)])
        output = transforms.Pad(padding)(img)
363
364
        self.assertEqual(output.size[0], width + padding[0] + padding[2])
        self.assertEqual(output.size[1], height + padding[1] + padding[3])
365

366
367
368
        # Checking if Padding can be printed as string
        transforms.Pad(padding).__repr__()

369
370
    def test_pad_with_non_constant_padding_modes(self):
        """Unit tests for edge, reflect, symmetric padding"""
vfdev's avatar
vfdev committed
371
        img = torch.zeros(3, 27, 27).byte()
372
373
374
375
376
377
378
379
380
        img[:, :, 0] = 1  # Constant value added to leftmost edge
        img = transforms.ToPILImage()(img)
        img = F.pad(img, 1, (200, 200, 200))

        # pad 3 to all sidess
        edge_padded_img = F.pad(img, 3, padding_mode='edge')
        # First 6 elements of leftmost edge in the middle of the image, values are in order:
        # edge_pad, edge_pad, edge_pad, constant_pad, constant value added to leftmost edge, 0
        edge_middle_slice = np.asarray(edge_padded_img).transpose(2, 0, 1)[0][17][:6]
381
382
        self.assertTrue(np.all(edge_middle_slice == np.asarray([200, 200, 200, 200, 1, 0])))
        self.assertEqual(transforms.ToTensor()(edge_padded_img).size(), (3, 35, 35))
383
384
385
386
387
388

        # Pad 3 to left/right, 2 to top/bottom
        reflect_padded_img = F.pad(img, (3, 2), padding_mode='reflect')
        # First 6 elements of leftmost edge in the middle of the image, values are in order:
        # reflect_pad, reflect_pad, reflect_pad, constant_pad, constant value added to leftmost edge, 0
        reflect_middle_slice = np.asarray(reflect_padded_img).transpose(2, 0, 1)[0][17][:6]
389
390
        self.assertTrue(np.all(reflect_middle_slice == np.asarray([0, 0, 1, 200, 1, 0])))
        self.assertEqual(transforms.ToTensor()(reflect_padded_img).size(), (3, 33, 35))
391
392
393
394
395
396

        # Pad 3 to left, 2 to top, 2 to right, 1 to bottom
        symmetric_padded_img = F.pad(img, (3, 2, 2, 1), padding_mode='symmetric')
        # First 6 elements of leftmost edge in the middle of the image, values are in order:
        # sym_pad, sym_pad, sym_pad, constant_pad, constant value added to leftmost edge, 0
        symmetric_middle_slice = np.asarray(symmetric_padded_img).transpose(2, 0, 1)[0][17][:6]
397
398
        self.assertTrue(np.all(symmetric_middle_slice == np.asarray([0, 1, 200, 200, 1, 0])))
        self.assertEqual(transforms.ToTensor()(symmetric_padded_img).size(), (3, 32, 34))
399

400
401
402
403
404
405
406
407
408
409
        # Check negative padding explicitly for symmetric case, since it is not
        # implemented for tensor case to compare to
        # Crop 1 to left, pad 2 to top, pad 3 to right, crop 3 to bottom
        symmetric_padded_img_neg = F.pad(img, (-1, 2, 3, -3), padding_mode='symmetric')
        symmetric_neg_middle_left = np.asarray(symmetric_padded_img_neg).transpose(2, 0, 1)[0][17][:3]
        symmetric_neg_middle_right = np.asarray(symmetric_padded_img_neg).transpose(2, 0, 1)[0][17][-4:]
        self.assertTrue(np.all(symmetric_neg_middle_left == np.asarray([1, 0, 0])))
        self.assertTrue(np.all(symmetric_neg_middle_right == np.asarray([200, 200, 0, 0])))
        self.assertEqual(transforms.ToTensor()(symmetric_padded_img_neg).size(), (3, 28, 31))

410
    def test_pad_raises_with_invalid_pad_sequence_len(self):
411
412
413
414
415
416
417
418
419
        with self.assertRaises(ValueError):
            transforms.Pad(())

        with self.assertRaises(ValueError):
            transforms.Pad((1, 2, 3))

        with self.assertRaises(ValueError):
            transforms.Pad((1, 2, 3, 4, 5))

420
421
422
423
424
425
426
427
    def test_pad_with_mode_F_images(self):
        pad = 2
        transform = transforms.Pad(pad)

        img = Image.new("F", (10, 10))
        padded_img = transform(img)
        self.assertSequenceEqual(padded_img.size, [edge_size + 2 * pad for edge_size in img.size])

Soumith Chintala's avatar
Soumith Chintala committed
428
429
430
431
    def test_lambda(self):
        trans = transforms.Lambda(lambda x: x.add(10))
        x = torch.randn(10)
        y = trans(x)
432
        self.assertTrue(y.equal(torch.add(x, 10)))
Soumith Chintala's avatar
Soumith Chintala committed
433
434
435
436

        trans = transforms.Lambda(lambda x: x.add_(10))
        x = torch.randn(10)
        y = trans(x)
437
        self.assertTrue(y.equal(x))
438

439
440
441
        # Checking if Lambda can be printed as string
        trans.__repr__()

442
    @unittest.skipIf(stats is None, 'scipy.stats not available')
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
    def test_random_apply(self):
        random_state = random.getstate()
        random.seed(42)
        random_apply_transform = transforms.RandomApply(
            [
                transforms.RandomRotation((-45, 45)),
                transforms.RandomHorizontalFlip(),
                transforms.RandomVerticalFlip(),
            ], p=0.75
        )
        img = transforms.ToPILImage()(torch.rand(3, 10, 10))
        num_samples = 250
        num_applies = 0
        for _ in range(num_samples):
            out = random_apply_transform(img)
            if out != img:
                num_applies += 1

        p_value = stats.binom_test(num_applies, num_samples, p=0.75)
        random.setstate(random_state)
463
        self.assertGreater(p_value, 0.0001)
464
465
466
467

        # Checking if RandomApply can be printed as string
        random_apply_transform.__repr__()

468
    @unittest.skipIf(stats is None, 'scipy.stats not available')
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
    def test_random_choice(self):
        random_state = random.getstate()
        random.seed(42)
        random_choice_transform = transforms.RandomChoice(
            [
                transforms.Resize(15),
                transforms.Resize(20),
                transforms.CenterCrop(10)
            ]
        )
        img = transforms.ToPILImage()(torch.rand(3, 25, 25))
        num_samples = 250
        num_resize_15 = 0
        num_resize_20 = 0
        num_crop_10 = 0
        for _ in range(num_samples):
            out = random_choice_transform(img)
            if out.size == (15, 15):
                num_resize_15 += 1
            elif out.size == (20, 20):
                num_resize_20 += 1
            elif out.size == (10, 10):
                num_crop_10 += 1

        p_value = stats.binom_test(num_resize_15, num_samples, p=0.33333)
494
        self.assertGreater(p_value, 0.0001)
495
        p_value = stats.binom_test(num_resize_20, num_samples, p=0.33333)
496
        self.assertGreater(p_value, 0.0001)
497
        p_value = stats.binom_test(num_crop_10, num_samples, p=0.33333)
498
        self.assertGreater(p_value, 0.0001)
499
500
501
502
503

        random.setstate(random_state)
        # Checking if RandomChoice can be printed as string
        random_choice_transform.__repr__()

504
    @unittest.skipIf(stats is None, 'scipy.stats not available')
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
    def test_random_order(self):
        random_state = random.getstate()
        random.seed(42)
        random_order_transform = transforms.RandomOrder(
            [
                transforms.Resize(20),
                transforms.CenterCrop(10)
            ]
        )
        img = transforms.ToPILImage()(torch.rand(3, 25, 25))
        num_samples = 250
        num_normal_order = 0
        resize_crop_out = transforms.CenterCrop(10)(transforms.Resize(20)(img))
        for _ in range(num_samples):
            out = random_order_transform(img)
            if out == resize_crop_out:
                num_normal_order += 1

        p_value = stats.binom_test(num_normal_order, num_samples, p=0.5)
        random.setstate(random_state)
525
        self.assertGreater(p_value, 0.0001)
526
527
528
529

        # Checking if RandomOrder can be printed as string
        random_order_transform.__repr__()

530
    def test_to_tensor(self):
531
        test_channels = [1, 3, 4]
532
533
        height, width = 4, 4
        trans = transforms.ToTensor()
534

535
536
537
538
539
540
541
        with self.assertRaises(TypeError):
            trans(np.random.rand(1, height, width).tolist())

        with self.assertRaises(ValueError):
            trans(np.random.rand(height))
            trans(np.random.rand(1, 1, height, width))

542
543
544
545
        for channels in test_channels:
            input_data = torch.ByteTensor(channels, height, width).random_(0, 255).float().div_(255)
            img = transforms.ToPILImage()(input_data)
            output = trans(img)
546
            self.assertTrue(np.allclose(input_data.numpy(), output.numpy()))
547

548
            ndarray = np.random.randint(low=0, high=255, size=(height, width, channels)).astype(np.uint8)
549
550
            output = trans(ndarray)
            expected_output = ndarray.transpose((2, 0, 1)) / 255.0
551
            self.assertTrue(np.allclose(output.numpy(), expected_output))
552

553
554
555
            ndarray = np.random.rand(height, width, channels).astype(np.float32)
            output = trans(ndarray)
            expected_output = ndarray.transpose((2, 0, 1))
556
            self.assertTrue(np.allclose(output.numpy(), expected_output))
557

558
559
560
561
        # separate test for mode '1' PIL images
        input_data = torch.ByteTensor(1, height, width).bernoulli_()
        img = transforms.ToPILImage()(input_data.mul(255)).convert('1')
        output = trans(img)
562
        self.assertTrue(np.allclose(input_data.numpy(), output.numpy()))
563

564
565
566
567
    def test_max_value(self):
        for dtype in int_dtypes():
            self.assertEqual(F_t._max_value(dtype), torch.iinfo(dtype).max)

568
569
570
571
        # remove float testing as it can lead to errors such as
        # runtime error: 5.7896e+76 is outside the range of representable values of type 'float'
        # for dtype in float_dtypes():
        #     self.assertGreater(F_t._max_value(dtype), torch.finfo(dtype).max)
572

573
574
575
576
577
578
    def test_convert_image_dtype_float_to_float(self):
        for input_dtype, output_dtypes in cycle_over(float_dtypes()):
            input_image = torch.tensor((0.0, 1.0), dtype=input_dtype)
            for output_dtype in output_dtypes:
                with self.subTest(input_dtype=input_dtype, output_dtype=output_dtype):
                    transform = transforms.ConvertImageDtype(output_dtype)
579
580
                    transform_script = torch.jit.script(F.convert_image_dtype)

581
                    output_image = transform(input_image)
582
583
584
585
                    output_image_script = transform_script(input_image, output_dtype)

                    script_diff = output_image_script - output_image
                    self.assertLess(script_diff.abs().max(), 1e-6)
586
587
588
589
590
591
592
593
594
595
596
597
598

                    actual_min, actual_max = output_image.tolist()
                    desired_min, desired_max = 0.0, 1.0

                    self.assertAlmostEqual(actual_min, desired_min)
                    self.assertAlmostEqual(actual_max, desired_max)

    def test_convert_image_dtype_float_to_int(self):
        for input_dtype in float_dtypes():
            input_image = torch.tensor((0.0, 1.0), dtype=input_dtype)
            for output_dtype in int_dtypes():
                with self.subTest(input_dtype=input_dtype, output_dtype=output_dtype):
                    transform = transforms.ConvertImageDtype(output_dtype)
599
                    transform_script = torch.jit.script(F.convert_image_dtype)
600
601
602
603
604
605
606
607

                    if (input_dtype == torch.float32 and output_dtype in (torch.int32, torch.int64)) or (
                            input_dtype == torch.float64 and output_dtype == torch.int64
                    ):
                        with self.assertRaises(RuntimeError):
                            transform(input_image)
                    else:
                        output_image = transform(input_image)
608
609
610
611
                        output_image_script = transform_script(input_image, output_dtype)

                        script_diff = output_image_script - output_image
                        self.assertLess(script_diff.abs().max(), 1e-6)
612
613
614
615
616
617
618
619
620
621
622
623
624

                        actual_min, actual_max = output_image.tolist()
                        desired_min, desired_max = 0, torch.iinfo(output_dtype).max

                        self.assertEqual(actual_min, desired_min)
                        self.assertEqual(actual_max, desired_max)

    def test_convert_image_dtype_int_to_float(self):
        for input_dtype in int_dtypes():
            input_image = torch.tensor((0, torch.iinfo(input_dtype).max), dtype=input_dtype)
            for output_dtype in float_dtypes():
                with self.subTest(input_dtype=input_dtype, output_dtype=output_dtype):
                    transform = transforms.ConvertImageDtype(output_dtype)
625
626
                    transform_script = torch.jit.script(F.convert_image_dtype)

627
                    output_image = transform(input_image)
628
629
630
631
                    output_image_script = transform_script(input_image, output_dtype)

                    script_diff = output_image_script - output_image
                    self.assertLess(script_diff.abs().max(), 1e-6)
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649

                    actual_min, actual_max = output_image.tolist()
                    desired_min, desired_max = 0.0, 1.0

                    self.assertAlmostEqual(actual_min, desired_min)
                    self.assertGreaterEqual(actual_min, desired_min)
                    self.assertAlmostEqual(actual_max, desired_max)
                    self.assertLessEqual(actual_max, desired_max)

    def test_convert_image_dtype_int_to_int(self):
        for input_dtype, output_dtypes in cycle_over(int_dtypes()):
            input_max = torch.iinfo(input_dtype).max
            input_image = torch.tensor((0, input_max), dtype=input_dtype)
            for output_dtype in output_dtypes:
                output_max = torch.iinfo(output_dtype).max

                with self.subTest(input_dtype=input_dtype, output_dtype=output_dtype):
                    transform = transforms.ConvertImageDtype(output_dtype)
650
651
                    transform_script = torch.jit.script(F.convert_image_dtype)

652
                    output_image = transform(input_image)
653
654
655
656
657
658
                    output_image_script = transform_script(input_image, output_dtype)

                    script_diff = output_image_script.float() - output_image.float()
                    self.assertLess(
                        script_diff.abs().max(), 1e-6, msg="{} vs {}".format(output_image_script, output_image)
                    )
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691

                    actual_min, actual_max = output_image.tolist()
                    desired_min, desired_max = 0, output_max

                    # see https://github.com/pytorch/vision/pull/2078#issuecomment-641036236 for details
                    if input_max >= output_max:
                        error_term = 0
                    else:
                        error_term = 1 - (torch.iinfo(output_dtype).max + 1) // (torch.iinfo(input_dtype).max + 1)

                    self.assertEqual(actual_min, desired_min)
                    self.assertEqual(actual_max, desired_max + error_term)

    def test_convert_image_dtype_int_to_int_consistency(self):
        for input_dtype, output_dtypes in cycle_over(int_dtypes()):
            input_max = torch.iinfo(input_dtype).max
            input_image = torch.tensor((0, input_max), dtype=input_dtype)
            for output_dtype in output_dtypes:
                output_max = torch.iinfo(output_dtype).max
                if output_max <= input_max:
                    continue

                with self.subTest(input_dtype=input_dtype, output_dtype=output_dtype):
                    transform = transforms.ConvertImageDtype(output_dtype)
                    inverse_transfrom = transforms.ConvertImageDtype(input_dtype)
                    output_image = inverse_transfrom(transform(input_image))

                    actual_min, actual_max = output_image.tolist()
                    desired_min, desired_max = 0, input_max

                    self.assertEqual(actual_min, desired_min)
                    self.assertEqual(actual_max, desired_max)

692
693
694
695
696
697
698
699
    @unittest.skipIf(accimage is None, 'accimage not available')
    def test_accimage_to_tensor(self):
        trans = transforms.ToTensor()

        expected_output = trans(Image.open(GRACE_HOPPER).convert('RGB'))
        output = trans(accimage.Image(GRACE_HOPPER))

        self.assertEqual(expected_output.size(), output.size())
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
        self.assertTrue(np.allclose(output.numpy(), expected_output.numpy()))

    def test_pil_to_tensor(self):
        test_channels = [1, 3, 4]
        height, width = 4, 4
        trans = transforms.PILToTensor()

        with self.assertRaises(TypeError):
            trans(np.random.rand(1, height, width).tolist())
            trans(np.random.rand(1, height, width))

        for channels in test_channels:
            input_data = torch.ByteTensor(channels, height, width).random_(0, 255)
            img = transforms.ToPILImage()(input_data)
            output = trans(img)
            self.assertTrue(np.allclose(input_data.numpy(), output.numpy()))

            input_data = np.random.randint(low=0, high=255, size=(height, width, channels)).astype(np.uint8)
            img = transforms.ToPILImage()(input_data)
            output = trans(img)
            expected_output = input_data.transpose((2, 0, 1))
            self.assertTrue(np.allclose(output.numpy(), expected_output))

            input_data = torch.as_tensor(np.random.rand(channels, height, width).astype(np.float32))
            img = transforms.ToPILImage()(input_data)  # CHW -> HWC and (* 255).byte()
            output = trans(img)  # HWC -> CHW
            expected_output = (input_data * 255).byte()
            self.assertTrue(np.allclose(output.numpy(), expected_output.numpy()))

        # separate test for mode '1' PIL images
        input_data = torch.ByteTensor(1, height, width).bernoulli_()
        img = transforms.ToPILImage()(input_data.mul(255)).convert('1')
        output = trans(img)
        self.assertTrue(np.allclose(input_data.numpy(), output.numpy()))

    @unittest.skipIf(accimage is None, 'accimage not available')
    def test_accimage_pil_to_tensor(self):
        trans = transforms.PILToTensor()

        expected_output = trans(Image.open(GRACE_HOPPER).convert('RGB'))
        output = trans(accimage.Image(GRACE_HOPPER))

        self.assertEqual(expected_output.size(), output.size())
743
        self.assertTrue(np.allclose(output.numpy(), expected_output.numpy()))
744
745
746
747

    @unittest.skipIf(accimage is None, 'accimage not available')
    def test_accimage_resize(self):
        trans = transforms.Compose([
748
            transforms.Resize(256, interpolation=Image.LINEAR),
749
750
751
            transforms.ToTensor(),
        ])

752
753
754
        # Checking if Compose, Resize and ToTensor can be printed as string
        trans.__repr__()

755
756
757
758
759
760
761
        expected_output = trans(Image.open(GRACE_HOPPER).convert('RGB'))
        output = trans(accimage.Image(GRACE_HOPPER))

        self.assertEqual(expected_output.size(), output.size())
        self.assertLess(np.abs((expected_output - output).mean()), 1e-3)
        self.assertLess((expected_output - output).var(), 1e-5)
        # note the high absolute tolerance
762
        self.assertTrue(np.allclose(output.numpy(), expected_output.numpy(), atol=5e-2))
763
764
765
766
767
768
769
770

    @unittest.skipIf(accimage is None, 'accimage not available')
    def test_accimage_crop(self):
        trans = transforms.Compose([
            transforms.CenterCrop(256),
            transforms.ToTensor(),
        ])

771
772
773
        # Checking if Compose, CenterCrop and ToTensor can be printed as string
        trans.__repr__()

774
775
776
777
        expected_output = trans(Image.open(GRACE_HOPPER).convert('RGB'))
        output = trans(accimage.Image(GRACE_HOPPER))

        self.assertEqual(expected_output.size(), output.size())
778
        self.assertTrue(np.allclose(output.numpy(), expected_output.numpy()))
779

780
    def test_1_channel_tensor_to_pil_image(self):
781
782
        to_tensor = transforms.ToTensor()

783
        img_data_float = torch.Tensor(1, 4, 4).uniform_()
784
785
786
787
        img_data_byte = torch.ByteTensor(1, 4, 4).random_(0, 255)
        img_data_short = torch.ShortTensor(1, 4, 4).random_()
        img_data_int = torch.IntTensor(1, 4, 4).random_()

788
789
790
791
792
793
794
795
796
797
        inputs = [img_data_float, img_data_byte, img_data_short, img_data_int]
        expected_outputs = [img_data_float.mul(255).int().float().div(255).numpy(),
                            img_data_byte.float().div(255.0).numpy(),
                            img_data_short.numpy(),
                            img_data_int.numpy()]
        expected_modes = ['L', 'L', 'I;16', 'I']

        for img_data, expected_output, mode in zip(inputs, expected_outputs, expected_modes):
            for transform in [transforms.ToPILImage(), transforms.ToPILImage(mode=mode)]:
                img = transform(img_data)
798
799
                self.assertEqual(img.mode, mode)
                self.assertTrue(np.allclose(expected_output, to_tensor(img).numpy()))
800
801
        # 'F' mode for torch.FloatTensor
        img_F_mode = transforms.ToPILImage(mode='F')(img_data_float)
802
803
804
        self.assertEqual(img_F_mode.mode, 'F')
        self.assertTrue(np.allclose(np.array(Image.fromarray(img_data_float.squeeze(0).numpy(), mode='F')),
                                    np.array(img_F_mode)))
805
806
807
808
809
810
811
812
813
814
815
816

    def test_1_channel_ndarray_to_pil_image(self):
        img_data_float = torch.Tensor(4, 4, 1).uniform_().numpy()
        img_data_byte = torch.ByteTensor(4, 4, 1).random_(0, 255).numpy()
        img_data_short = torch.ShortTensor(4, 4, 1).random_().numpy()
        img_data_int = torch.IntTensor(4, 4, 1).random_().numpy()

        inputs = [img_data_float, img_data_byte, img_data_short, img_data_int]
        expected_modes = ['F', 'L', 'I;16', 'I']
        for img_data, mode in zip(inputs, expected_modes):
            for transform in [transforms.ToPILImage(), transforms.ToPILImage(mode=mode)]:
                img = transform(img_data)
817
818
                self.assertEqual(img.mode, mode)
                self.assertTrue(np.allclose(img_data[:, :, 0], img))
819

surgan12's avatar
surgan12 committed
820
821
822
823
    def test_2_channel_ndarray_to_pil_image(self):
        def verify_img_data(img_data, mode):
            if mode is None:
                img = transforms.ToPILImage()(img_data)
824
                self.assertEqual(img.mode, 'LA')  # default should assume LA
surgan12's avatar
surgan12 committed
825
826
            else:
                img = transforms.ToPILImage(mode=mode)(img_data)
827
                self.assertEqual(img.mode, mode)
surgan12's avatar
surgan12 committed
828
829
            split = img.split()
            for i in range(2):
830
                self.assertTrue(np.allclose(img_data[:, :, i], split[i]))
surgan12's avatar
surgan12 committed
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847

        img_data = torch.ByteTensor(4, 4, 2).random_(0, 255).numpy()
        for mode in [None, 'LA']:
            verify_img_data(img_data, mode)

        transforms.ToPILImage().__repr__()

        with self.assertRaises(ValueError):
            # should raise if we try a mode for 4 or 1 or 3 channel images
            transforms.ToPILImage(mode='RGBA')(img_data)
            transforms.ToPILImage(mode='P')(img_data)
            transforms.ToPILImage(mode='RGB')(img_data)

    def test_2_channel_tensor_to_pil_image(self):
        def verify_img_data(img_data, expected_output, mode):
            if mode is None:
                img = transforms.ToPILImage()(img_data)
848
                self.assertEqual(img.mode, 'LA')  # default should assume LA
surgan12's avatar
surgan12 committed
849
850
            else:
                img = transforms.ToPILImage(mode=mode)(img_data)
851
                self.assertEqual(img.mode, mode)
surgan12's avatar
surgan12 committed
852
853
            split = img.split()
            for i in range(2):
854
                self.assertTrue(np.allclose(expected_output[i].numpy(), F.to_tensor(split[i]).numpy()))
surgan12's avatar
surgan12 committed
855
856
857
858
859
860
861
862
863
864
865
866

        img_data = torch.Tensor(2, 4, 4).uniform_()
        expected_output = img_data.mul(255).int().float().div(255)
        for mode in [None, 'LA']:
            verify_img_data(img_data, expected_output, mode=mode)

        with self.assertRaises(ValueError):
            # should raise if we try a mode for 4 or 1 or 3 channel images
            transforms.ToPILImage(mode='RGBA')(img_data)
            transforms.ToPILImage(mode='P')(img_data)
            transforms.ToPILImage(mode='RGB')(img_data)

867
868
869
870
    def test_3_channel_tensor_to_pil_image(self):
        def verify_img_data(img_data, expected_output, mode):
            if mode is None:
                img = transforms.ToPILImage()(img_data)
871
                self.assertEqual(img.mode, 'RGB')  # default should assume RGB
872
873
            else:
                img = transforms.ToPILImage(mode=mode)(img_data)
874
                self.assertEqual(img.mode, mode)
875
876
            split = img.split()
            for i in range(3):
877
                self.assertTrue(np.allclose(expected_output[i].numpy(), F.to_tensor(split[i]).numpy()))
878

879
880
881
882
        img_data = torch.Tensor(3, 4, 4).uniform_()
        expected_output = img_data.mul(255).int().float().div(255)
        for mode in [None, 'RGB', 'HSV', 'YCbCr']:
            verify_img_data(img_data, expected_output, mode=mode)
883

884
        with self.assertRaises(ValueError):
surgan12's avatar
surgan12 committed
885
            # should raise if we try a mode for 4 or 1 or 2 channel images
886
887
            transforms.ToPILImage(mode='RGBA')(img_data)
            transforms.ToPILImage(mode='P')(img_data)
surgan12's avatar
surgan12 committed
888
            transforms.ToPILImage(mode='LA')(img_data)
889

Varun Agrawal's avatar
Varun Agrawal committed
890
891
892
        with self.assertRaises(ValueError):
            transforms.ToPILImage()(torch.Tensor(1, 3, 4, 4).uniform_())

893
894
895
896
    def test_3_channel_ndarray_to_pil_image(self):
        def verify_img_data(img_data, mode):
            if mode is None:
                img = transforms.ToPILImage()(img_data)
897
                self.assertEqual(img.mode, 'RGB')  # default should assume RGB
898
899
            else:
                img = transforms.ToPILImage(mode=mode)(img_data)
900
                self.assertEqual(img.mode, mode)
901
902
            split = img.split()
            for i in range(3):
903
                self.assertTrue(np.allclose(img_data[:, :, i], split[i]))
904

905
906
907
908
        img_data = torch.ByteTensor(4, 4, 3).random_(0, 255).numpy()
        for mode in [None, 'RGB', 'HSV', 'YCbCr']:
            verify_img_data(img_data, mode)

909
910
911
        # Checking if ToPILImage can be printed as string
        transforms.ToPILImage().__repr__()

912
        with self.assertRaises(ValueError):
surgan12's avatar
surgan12 committed
913
            # should raise if we try a mode for 4 or 1 or 2 channel images
914
915
            transforms.ToPILImage(mode='RGBA')(img_data)
            transforms.ToPILImage(mode='P')(img_data)
surgan12's avatar
surgan12 committed
916
            transforms.ToPILImage(mode='LA')(img_data)
917
918
919
920
921

    def test_4_channel_tensor_to_pil_image(self):
        def verify_img_data(img_data, expected_output, mode):
            if mode is None:
                img = transforms.ToPILImage()(img_data)
922
                self.assertEqual(img.mode, 'RGBA')  # default should assume RGBA
923
924
            else:
                img = transforms.ToPILImage(mode=mode)(img_data)
925
                self.assertEqual(img.mode, mode)
926
927
928

            split = img.split()
            for i in range(4):
929
                self.assertTrue(np.allclose(expected_output[i].numpy(), F.to_tensor(split[i]).numpy()))
930

931
        img_data = torch.Tensor(4, 4, 4).uniform_()
932
        expected_output = img_data.mul(255).int().float().div(255)
surgan12's avatar
surgan12 committed
933
        for mode in [None, 'RGBA', 'CMYK', 'RGBX']:
934
            verify_img_data(img_data, expected_output, mode)
935

936
        with self.assertRaises(ValueError):
surgan12's avatar
surgan12 committed
937
            # should raise if we try a mode for 3 or 1 or 2 channel images
938
939
            transforms.ToPILImage(mode='RGB')(img_data)
            transforms.ToPILImage(mode='P')(img_data)
surgan12's avatar
surgan12 committed
940
            transforms.ToPILImage(mode='LA')(img_data)
941
942
943
944
945

    def test_4_channel_ndarray_to_pil_image(self):
        def verify_img_data(img_data, mode):
            if mode is None:
                img = transforms.ToPILImage()(img_data)
946
                self.assertEqual(img.mode, 'RGBA')  # default should assume RGBA
947
948
            else:
                img = transforms.ToPILImage(mode=mode)(img_data)
949
                self.assertEqual(img.mode, mode)
950
951
            split = img.split()
            for i in range(4):
952
                self.assertTrue(np.allclose(img_data[:, :, i], split[i]))
953

954
        img_data = torch.ByteTensor(4, 4, 4).random_(0, 255).numpy()
surgan12's avatar
surgan12 committed
955
        for mode in [None, 'RGBA', 'CMYK', 'RGBX']:
956
            verify_img_data(img_data, mode)
957

958
        with self.assertRaises(ValueError):
surgan12's avatar
surgan12 committed
959
            # should raise if we try a mode for 3 or 1 or 2 channel images
960
961
            transforms.ToPILImage(mode='RGB')(img_data)
            transforms.ToPILImage(mode='P')(img_data)
surgan12's avatar
surgan12 committed
962
            transforms.ToPILImage(mode='LA')(img_data)
963

Varun Agrawal's avatar
Varun Agrawal committed
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
    def test_2d_tensor_to_pil_image(self):
        to_tensor = transforms.ToTensor()

        img_data_float = torch.Tensor(4, 4).uniform_()
        img_data_byte = torch.ByteTensor(4, 4).random_(0, 255)
        img_data_short = torch.ShortTensor(4, 4).random_()
        img_data_int = torch.IntTensor(4, 4).random_()

        inputs = [img_data_float, img_data_byte, img_data_short, img_data_int]
        expected_outputs = [img_data_float.mul(255).int().float().div(255).numpy(),
                            img_data_byte.float().div(255.0).numpy(),
                            img_data_short.numpy(),
                            img_data_int.numpy()]
        expected_modes = ['L', 'L', 'I;16', 'I']

        for img_data, expected_output, mode in zip(inputs, expected_outputs, expected_modes):
            for transform in [transforms.ToPILImage(), transforms.ToPILImage(mode=mode)]:
                img = transform(img_data)
982
983
                self.assertEqual(img.mode, mode)
                self.assertTrue(np.allclose(expected_output, to_tensor(img).numpy()))
Varun Agrawal's avatar
Varun Agrawal committed
984
985
986
987
988
989
990
991
992
993
994
995

    def test_2d_ndarray_to_pil_image(self):
        img_data_float = torch.Tensor(4, 4).uniform_().numpy()
        img_data_byte = torch.ByteTensor(4, 4).random_(0, 255).numpy()
        img_data_short = torch.ShortTensor(4, 4).random_().numpy()
        img_data_int = torch.IntTensor(4, 4).random_().numpy()

        inputs = [img_data_float, img_data_byte, img_data_short, img_data_int]
        expected_modes = ['F', 'L', 'I;16', 'I']
        for img_data, mode in zip(inputs, expected_modes):
            for transform in [transforms.ToPILImage(), transforms.ToPILImage(mode=mode)]:
                img = transform(img_data)
996
997
                self.assertEqual(img.mode, mode)
                self.assertTrue(np.allclose(img_data, img))
Varun Agrawal's avatar
Varun Agrawal committed
998
999

    def test_tensor_bad_types_to_pil_image(self):
1000
        with self.assertRaisesRegex(ValueError, r'pic should be 2/3 dimensional. Got \d+ dimensions.'):
Varun Agrawal's avatar
Varun Agrawal committed
1001
            transforms.ToPILImage()(torch.ones(1, 3, 4, 4))
1002
1003
        with self.assertRaisesRegex(ValueError, r'pic should not have > 4 channels. Got \d+ channels.'):
            transforms.ToPILImage()(torch.ones(6, 4, 4))
Varun Agrawal's avatar
Varun Agrawal committed
1004

1005
    def test_ndarray_bad_types_to_pil_image(self):
1006
        trans = transforms.ToPILImage()
1007
1008
        reg_msg = r'Input type \w+ is not supported'
        with self.assertRaisesRegex(TypeError, reg_msg):
1009
            trans(np.ones([4, 4, 1], np.int64))
1010
        with self.assertRaisesRegex(TypeError, reg_msg):
1011
            trans(np.ones([4, 4, 1], np.uint16))
1012
        with self.assertRaisesRegex(TypeError, reg_msg):
1013
            trans(np.ones([4, 4, 1], np.uint32))
1014
        with self.assertRaisesRegex(TypeError, reg_msg):
1015
1016
            trans(np.ones([4, 4, 1], np.float64))

1017
        with self.assertRaisesRegex(ValueError, r'pic should be 2/3 dimensional. Got \d+ dimensions.'):
Varun Agrawal's avatar
Varun Agrawal committed
1018
            transforms.ToPILImage()(np.ones([1, 4, 4, 3]))
1019
1020
        with self.assertRaisesRegex(ValueError, r'pic should not have > 4 channels. Got \d+ channels.'):
            transforms.ToPILImage()(np.ones([4, 4, 6]))
Varun Agrawal's avatar
Varun Agrawal committed
1021

1022
1023
    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_vertical_flip(self):
1024
1025
        random_state = random.getstate()
        random.seed(42)
1026
1027
1028
        img = transforms.ToPILImage()(torch.rand(3, 10, 10))
        vimg = img.transpose(Image.FLIP_TOP_BOTTOM)

1029
        num_samples = 250
1030
        num_vertical = 0
1031
        for _ in range(num_samples):
1032
1033
1034
1035
            out = transforms.RandomVerticalFlip()(img)
            if out == vimg:
                num_vertical += 1

1036
1037
        p_value = stats.binom_test(num_vertical, num_samples, p=0.5)
        random.setstate(random_state)
1038
        self.assertGreater(p_value, 0.0001)
1039

1040
1041
1042
1043
1044
1045
1046
1047
1048
        num_samples = 250
        num_vertical = 0
        for _ in range(num_samples):
            out = transforms.RandomVerticalFlip(p=0.7)(img)
            if out == vimg:
                num_vertical += 1

        p_value = stats.binom_test(num_vertical, num_samples, p=0.7)
        random.setstate(random_state)
1049
        self.assertGreater(p_value, 0.0001)
1050

1051
1052
1053
        # Checking if RandomVerticalFlip can be printed as string
        transforms.RandomVerticalFlip().__repr__()

1054
1055
    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_horizontal_flip(self):
1056
1057
        random_state = random.getstate()
        random.seed(42)
1058
1059
1060
        img = transforms.ToPILImage()(torch.rand(3, 10, 10))
        himg = img.transpose(Image.FLIP_LEFT_RIGHT)

1061
        num_samples = 250
1062
        num_horizontal = 0
1063
        for _ in range(num_samples):
1064
1065
1066
1067
            out = transforms.RandomHorizontalFlip()(img)
            if out == himg:
                num_horizontal += 1

1068
1069
        p_value = stats.binom_test(num_horizontal, num_samples, p=0.5)
        random.setstate(random_state)
1070
        self.assertGreater(p_value, 0.0001)
1071

1072
1073
1074
1075
1076
1077
1078
1079
1080
        num_samples = 250
        num_horizontal = 0
        for _ in range(num_samples):
            out = transforms.RandomHorizontalFlip(p=0.7)(img)
            if out == himg:
                num_horizontal += 1

        p_value = stats.binom_test(num_horizontal, num_samples, p=0.7)
        random.setstate(random_state)
1081
        self.assertGreater(p_value, 0.0001)
1082

1083
1084
1085
        # Checking if RandomHorizontalFlip can be printed as string
        transforms.RandomHorizontalFlip().__repr__()

1086
    @unittest.skipIf(stats is None, 'scipy.stats is not available')
1087
1088
1089
1090
1091
1092
1093
1094
1095
1096
1097
1098
    def test_normalize(self):
        def samples_from_standard_normal(tensor):
            p_value = stats.kstest(list(tensor.view(-1)), 'norm', args=(0, 1)).pvalue
            return p_value > 0.0001

        random_state = random.getstate()
        random.seed(42)
        for channels in [1, 3]:
            img = torch.rand(channels, 10, 10)
            mean = [img[c].mean() for c in range(channels)]
            std = [img[c].std() for c in range(channels)]
            normalized = transforms.Normalize(mean, std)(img)
1099
            self.assertTrue(samples_from_standard_normal(normalized))
1100
1101
        random.setstate(random_state)

1102
1103
1104
        # Checking if Normalize can be printed as string
        transforms.Normalize(mean, std).__repr__()

1105
1106
1107
        # Checking the optional in-place behaviour
        tensor = torch.rand((1, 16, 16))
        tensor_inplace = transforms.Normalize((0.5,), (0.5,), inplace=True)(tensor)
1108
        self.assertTrue(torch.equal(tensor, tensor_inplace))
1109

1110
1111
1112
1113
1114
1115
1116
1117
1118
    def test_normalize_different_dtype(self):
        for dtype1 in [torch.float32, torch.float64]:
            img = torch.rand(3, 10, 10, dtype=dtype1)
            for dtype2 in [torch.int64, torch.float32, torch.float64]:
                mean = torch.tensor([1, 2, 3], dtype=dtype2)
                std = torch.tensor([1, 2, 1], dtype=dtype2)
                # checks that it doesn't crash
                transforms.functional.normalize(img, mean, std)

1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131
1132
1133
1134
1135
1136
    def test_normalize_3d_tensor(self):
        torch.manual_seed(28)
        n_channels = 3
        img_size = 10
        mean = torch.rand(n_channels)
        std = torch.rand(n_channels)
        img = torch.rand(n_channels, img_size, img_size)
        target = F.normalize(img, mean, std).numpy()

        mean_unsqueezed = mean.view(-1, 1, 1)
        std_unsqueezed = std.view(-1, 1, 1)
        result1 = F.normalize(img, mean_unsqueezed, std_unsqueezed)
        result2 = F.normalize(img,
                              mean_unsqueezed.repeat(1, img_size, img_size),
                              std_unsqueezed.repeat(1, img_size, img_size))
        assert_array_almost_equal(target, result1.numpy())
        assert_array_almost_equal(target, result2.numpy())

1137
1138
1139
1140
1141
1142
1143
    def test_adjust_brightness(self):
        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')

        # test 0
1144
        y_pil = F.adjust_brightness(x_pil, 1)
1145
        y_np = np.array(y_pil)
1146
        self.assertTrue(np.allclose(y_np, x_np))
1147
1148

        # test 1
1149
        y_pil = F.adjust_brightness(x_pil, 0.5)
1150
1151
1152
        y_np = np.array(y_pil)
        y_ans = [0, 2, 6, 27, 67, 113, 18, 4, 117, 45, 127, 0]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1153
        self.assertTrue(np.allclose(y_np, y_ans))
1154
1155

        # test 2
1156
        y_pil = F.adjust_brightness(x_pil, 2)
1157
1158
1159
        y_np = np.array(y_pil)
        y_ans = [0, 10, 26, 108, 255, 255, 74, 16, 255, 180, 255, 2]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1160
        self.assertTrue(np.allclose(y_np, y_ans))
1161
1162
1163
1164
1165
1166
1167
1168

    def test_adjust_contrast(self):
        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')

        # test 0
1169
        y_pil = F.adjust_contrast(x_pil, 1)
1170
        y_np = np.array(y_pil)
1171
        self.assertTrue(np.allclose(y_np, x_np))
1172
1173

        # test 1
1174
        y_pil = F.adjust_contrast(x_pil, 0.5)
1175
1176
1177
        y_np = np.array(y_pil)
        y_ans = [43, 45, 49, 70, 110, 156, 61, 47, 160, 88, 170, 43]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1178
        self.assertTrue(np.allclose(y_np, y_ans))
1179
1180

        # test 2
1181
        y_pil = F.adjust_contrast(x_pil, 2)
1182
1183
1184
        y_np = np.array(y_pil)
        y_ans = [0, 0, 0, 22, 184, 255, 0, 0, 255, 94, 255, 0]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1185
        self.assertTrue(np.allclose(y_np, y_ans))
1186

Francisco Massa's avatar
Francisco Massa committed
1187
    @unittest.skipIf(Image.__version__ >= '7', "Temporarily disabled")
1188
1189
1190
1191
1192
1193
1194
    def test_adjust_saturation(self):
        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')

        # test 0
1195
        y_pil = F.adjust_saturation(x_pil, 1)
1196
        y_np = np.array(y_pil)
1197
        self.assertTrue(np.allclose(y_np, x_np))
1198
1199

        # test 1
1200
        y_pil = F.adjust_saturation(x_pil, 0.5)
1201
1202
1203
        y_np = np.array(y_pil)
        y_ans = [2, 4, 8, 87, 128, 173, 39, 25, 138, 133, 215, 88]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1204
        self.assertTrue(np.allclose(y_np, y_ans))
1205
1206

        # test 2
1207
        y_pil = F.adjust_saturation(x_pil, 2)
1208
1209
1210
        y_np = np.array(y_pil)
        y_ans = [0, 6, 22, 0, 149, 255, 32, 0, 255, 4, 255, 0]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1211
        self.assertTrue(np.allclose(y_np, y_ans))
1212
1213
1214
1215
1216
1217
1218
1219

    def test_adjust_hue(self):
        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')

        with self.assertRaises(ValueError):
1220
1221
            F.adjust_hue(x_pil, -0.7)
            F.adjust_hue(x_pil, 1)
1222
1223
1224

        # test 0: almost same as x_data but not exact.
        # probably because hsv <-> rgb floating point ops
1225
        y_pil = F.adjust_hue(x_pil, 0)
1226
1227
1228
        y_np = np.array(y_pil)
        y_ans = [0, 5, 13, 54, 139, 226, 35, 8, 234, 91, 255, 1]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1229
        self.assertTrue(np.allclose(y_np, y_ans))
1230
1231

        # test 1
1232
        y_pil = F.adjust_hue(x_pil, 0.25)
1233
1234
1235
        y_np = np.array(y_pil)
        y_ans = [13, 0, 12, 224, 54, 226, 234, 8, 99, 1, 222, 255]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1236
        self.assertTrue(np.allclose(y_np, y_ans))
1237
1238

        # test 2
1239
        y_pil = F.adjust_hue(x_pil, -0.25)
1240
1241
1242
        y_np = np.array(y_pil)
        y_ans = [0, 13, 2, 54, 226, 58, 8, 234, 152, 255, 43, 1]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1243
        self.assertTrue(np.allclose(y_np, y_ans))
1244

1245
1246
1247
1248
1249
1250
1251
1252
1253
1254
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
1274
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
    def test_adjust_sharpness(self):
        x_shape = [4, 4, 3]
        x_data = [75, 121, 114, 105, 97, 107, 105, 32, 66, 111, 117, 114, 99, 104, 97, 0,
                  0, 65, 108, 101, 120, 97, 110, 100, 101, 114, 32, 86, 114, 121, 110, 105,
                  111, 116, 105, 115, 0, 0, 73, 32, 108, 111, 118, 101, 32, 121, 111, 117]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')

        # test 0
        y_pil = F.adjust_sharpness(x_pil, 1)
        y_np = np.array(y_pil)
        self.assertTrue(np.allclose(y_np, x_np))

        # test 1
        y_pil = F.adjust_sharpness(x_pil, 0.5)
        y_np = np.array(y_pil)
        y_ans = [75, 121, 114, 105, 97, 107, 105, 32, 66, 111, 117, 114, 99, 104, 97, 30,
                 30, 74, 103, 96, 114, 97, 110, 100, 101, 114, 32, 81, 103, 108, 102, 101,
                 107, 116, 105, 115, 0, 0, 73, 32, 108, 111, 118, 101, 32, 121, 111, 117]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
        self.assertTrue(np.allclose(y_np, y_ans))

        # test 2
        y_pil = F.adjust_sharpness(x_pil, 2)
        y_np = np.array(y_pil)
        y_ans = [75, 121, 114, 105, 97, 107, 105, 32, 66, 111, 117, 114, 99, 104, 97, 0,
                 0, 46, 118, 111, 132, 97, 110, 100, 101, 114, 32, 95, 135, 146, 126, 112,
                 119, 116, 105, 115, 0, 0, 73, 32, 108, 111, 118, 101, 32, 121, 111, 117]
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
        self.assertTrue(np.allclose(y_np, y_ans))

        # test 3
        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')
        x_th = torch.tensor(x_np.transpose(2, 0, 1))
        y_pil = F.adjust_sharpness(x_pil, 2)
        y_np = np.array(y_pil).transpose(2, 0, 1)
        y_th = F.adjust_sharpness(x_th, 2)
        self.assertTrue(np.allclose(y_np, y_th.numpy()))

1287
1288
1289
1290
1291
1292
1293
    def test_adjust_gamma(self):
        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')

        # test 0
1294
        y_pil = F.adjust_gamma(x_pil, 1)
1295
        y_np = np.array(y_pil)
1296
        self.assertTrue(np.allclose(y_np, x_np))
1297
1298

        # test 1
1299
        y_pil = F.adjust_gamma(x_pil, 0.5)
1300
        y_np = np.array(y_pil)
1301
        y_ans = [0, 35, 57, 117, 186, 241, 97, 45, 245, 152, 255, 16]
1302
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1303
        self.assertTrue(np.allclose(y_np, y_ans))
1304
1305

        # test 2
1306
        y_pil = F.adjust_gamma(x_pil, 2)
1307
        y_np = np.array(y_pil)
1308
        y_ans = [0, 0, 0, 11, 71, 201, 5, 0, 215, 31, 255, 0]
1309
        y_ans = np.array(y_ans, dtype=np.uint8).reshape(x_shape)
1310
        self.assertTrue(np.allclose(y_np, y_ans))
1311
1312
1313
1314
1315
1316
1317
1318

    def test_adjusts_L_mode(self):
        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_rgb = Image.fromarray(x_np, mode='RGB')

        x_l = x_rgb.convert('L')
1319
1320
1321
1322
        self.assertEqual(F.adjust_brightness(x_l, 2).mode, 'L')
        self.assertEqual(F.adjust_saturation(x_l, 2).mode, 'L')
        self.assertEqual(F.adjust_contrast(x_l, 2).mode, 'L')
        self.assertEqual(F.adjust_hue(x_l, 0.4).mode, 'L')
1323
        self.assertEqual(F.adjust_sharpness(x_l, 2).mode, 'L')
1324
        self.assertEqual(F.adjust_gamma(x_l, 0.5).mode, 'L')
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336

    def test_color_jitter(self):
        color_jitter = transforms.ColorJitter(2, 2, 2, 0.1)

        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')
        x_pil_2 = x_pil.convert('L')

        for i in range(10):
            y_pil = color_jitter(x_pil)
1337
            self.assertEqual(y_pil.mode, x_pil.mode)
1338
1339

            y_pil_2 = color_jitter(x_pil_2)
1340
            self.assertEqual(y_pil_2.mode, x_pil_2.mode)
1341

1342
1343
1344
        # Checking if ColorJitter can be printed as string
        color_jitter.__repr__()

1345
    def test_linear_transformation(self):
ekka's avatar
ekka committed
1346
1347
1348
1349
1350
1351
1352
1353
1354
1355
1356
1357
        num_samples = 1000
        x = torch.randn(num_samples, 3, 10, 10)
        flat_x = x.view(x.size(0), x.size(1) * x.size(2) * x.size(3))
        # compute principal components
        sigma = torch.mm(flat_x.t(), flat_x) / flat_x.size(0)
        u, s, _ = np.linalg.svd(sigma.numpy())
        zca_epsilon = 1e-10  # avoid division by 0
        d = torch.Tensor(np.diag(1. / np.sqrt(s + zca_epsilon)))
        u = torch.Tensor(u)
        principal_components = torch.mm(torch.mm(u, d), u.t())
        mean_vector = (torch.sum(flat_x, dim=0) / flat_x.size(0))
        # initialize whitening matrix
1358
        whitening = transforms.LinearTransformation(principal_components, mean_vector)
ekka's avatar
ekka committed
1359
1360
1361
1362
1363
1364
1365
1366
1367
1368
        # estimate covariance and mean using weak law of large number
        num_features = flat_x.size(1)
        cov = 0.0
        mean = 0.0
        for i in x:
            xwhite = whitening(i)
            xwhite = xwhite.view(1, -1).numpy()
            cov += np.dot(xwhite, xwhite.T) / num_features
            mean += np.sum(xwhite) / num_features
        # if rtol for std = 1e-3 then rtol for cov = 2e-3 as std**2 = cov
1369
1370
1371
1372
        self.assertTrue(np.allclose(cov / num_samples, np.identity(1), rtol=2e-3),
                        "cov not close to 1")
        self.assertTrue(np.allclose(mean / num_samples, 0, rtol=1e-3),
                        "mean not close to 0")
ekka's avatar
ekka committed
1373

1374
        # Checking if LinearTransformation can be printed as string
ekka's avatar
ekka committed
1375
1376
        whitening.__repr__()

1377
1378
1379
1380
    def test_rotate(self):
        x = np.zeros((100, 100, 3), dtype=np.uint8)
        x[40, 40] = [255, 255, 255]

vfdev's avatar
vfdev committed
1381
        with self.assertRaisesRegex(TypeError, r"img should be PIL Image"):
1382
1383
1384
1385
1386
            F.rotate(x, 10)

        img = F.to_pil_image(x)

        result = F.rotate(img, 45)
1387
        self.assertEqual(result.size, (100, 100))
1388
        r, c, ch = np.where(result)
1389
1390
1391
        self.assertTrue(all(x in r for x in [49, 50]))
        self.assertTrue(all(x in c for x in [36]))
        self.assertTrue(all(x in ch for x in [0, 1, 2]))
1392
1393

        result = F.rotate(img, 45, expand=True)
1394
        self.assertEqual(result.size, (142, 142))
1395
        r, c, ch = np.where(result)
1396
1397
1398
        self.assertTrue(all(x in r for x in [70, 71]))
        self.assertTrue(all(x in c for x in [57]))
        self.assertTrue(all(x in ch for x in [0, 1, 2]))
1399
1400

        result = F.rotate(img, 45, center=(40, 40))
1401
        self.assertEqual(result.size, (100, 100))
1402
        r, c, ch = np.where(result)
1403
1404
1405
        self.assertTrue(all(x in r for x in [40]))
        self.assertTrue(all(x in c for x in [40]))
        self.assertTrue(all(x in ch for x in [0, 1, 2]))
1406
1407
1408
1409

        result_a = F.rotate(img, 90)
        result_b = F.rotate(img, -270)

1410
        self.assertTrue(np.all(np.array(result_a) == np.array(result_b)))
1411

Philip Meier's avatar
Philip Meier committed
1412
1413
1414
    def test_rotate_fill(self):
        img = F.to_pil_image(np.ones((100, 100, 3), dtype=np.uint8) * 255, "RGB")

1415
        modes = ("L", "RGB", "F")
Philip Meier's avatar
Philip Meier committed
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429
1430
1431
        nums_bands = [len(mode) for mode in modes]
        fill = 127

        for mode, num_bands in zip(modes, nums_bands):
            img_conv = img.convert(mode)
            img_rot = F.rotate(img_conv, 45.0, fill=fill)
            pixel = img_rot.getpixel((0, 0))

            if not isinstance(pixel, tuple):
                pixel = (pixel,)
            self.assertTupleEqual(pixel, tuple([fill] * num_bands))

            for wrong_num_bands in set(nums_bands) - {num_bands}:
                with self.assertRaises(ValueError):
                    F.rotate(img_conv, 45.0, fill=tuple([fill] * wrong_num_bands))

1432
    def test_affine(self):
Francisco Massa's avatar
Francisco Massa committed
1433
1434
1435
        input_img = np.zeros((40, 40, 3), dtype=np.uint8)
        cnt = [20, 20]
        for pt in [(16, 16), (20, 16), (20, 20)]:
1436
1437
1438
1439
            for i in range(-5, 5):
                for j in range(-5, 5):
                    input_img[pt[0] + i, pt[1] + j, :] = [255, 155, 55]

vfdev's avatar
vfdev committed
1440
1441
        with self.assertRaises(TypeError, msg="Argument translate should be a sequence"):
            F.affine(input_img, 10, translate=0, scale=1, shear=1)
1442
1443
1444
1445
1446
1447
1448
1449
1450
1451
1452

        pil_img = F.to_pil_image(input_img)

        def _to_3x3_inv(inv_result_matrix):
            result_matrix = np.zeros((3, 3))
            result_matrix[:2, :] = np.array(inv_result_matrix).reshape((2, 3))
            result_matrix[2, 2] = 1
            return np.linalg.inv(result_matrix)

        def _test_transformation(a, t, s, sh):
            a_rad = math.radians(a)
ptrblck's avatar
ptrblck committed
1453
            s_rad = [math.radians(sh_) for sh_ in sh]
1454
1455
1456
1457
1458
            cx, cy = cnt
            tx, ty = t
            sx, sy = s_rad
            rot = a_rad

1459
            # 1) Check transformation matrix:
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470
1471
1472
1473
1474
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
            C = np.array([[1, 0, cx],
                          [0, 1, cy],
                          [0, 0, 1]])
            T = np.array([[1, 0, tx],
                          [0, 1, ty],
                          [0, 0, 1]])
            Cinv = np.linalg.inv(C)

            RS = np.array(
                [[s * math.cos(rot), -s * math.sin(rot), 0],
                 [s * math.sin(rot), s * math.cos(rot), 0],
                 [0, 0, 1]])

            SHx = np.array([[1, -math.tan(sx), 0],
                            [0, 1, 0],
                            [0, 0, 1]])

            SHy = np.array([[1, 0, 0],
                            [-math.tan(sy), 1, 0],
                            [0, 0, 1]])

            RSS = np.matmul(RS, np.matmul(SHy, SHx))

            true_matrix = np.matmul(T, np.matmul(C, np.matmul(RSS, Cinv)))

1485
1486
            result_matrix = _to_3x3_inv(F._get_inverse_affine_matrix(center=cnt, angle=a,
                                                                     translate=t, scale=s, shear=sh))
1487
            self.assertLess(np.sum(np.abs(true_matrix - result_matrix)), 1e-10)
1488
            # 2) Perform inverse mapping:
Francisco Massa's avatar
Francisco Massa committed
1489
            true_result = np.zeros((40, 40, 3), dtype=np.uint8)
1490
1491
1492
            inv_true_matrix = np.linalg.inv(true_matrix)
            for y in range(true_result.shape[0]):
                for x in range(true_result.shape[1]):
1493
1494
1495
1496
1497
1498
                    # Same as for PIL:
                    # https://github.com/python-pillow/Pillow/blob/71f8ec6a0cfc1008076a023c0756542539d057ab/
                    # src/libImaging/Geometry.c#L1060
                    input_pt = np.array([x + 0.5, y + 0.5, 1.0])
                    res = np.floor(np.dot(inv_true_matrix, input_pt)).astype(np.int)
                    _x, _y = res[:2]
1499
1500
1501
1502
                    if 0 <= _x < input_img.shape[1] and 0 <= _y < input_img.shape[0]:
                        true_result[y, x, :] = input_img[_y, _x, :]

            result = F.affine(pil_img, angle=a, translate=t, scale=s, shear=sh)
1503
            self.assertEqual(result.size, pil_img.size)
1504
1505
1506
1507
            # Compute number of different pixels:
            np_result = np.array(result)
            n_diff_pixels = np.sum(np_result != true_result) / 3
            # Accept 3 wrong pixels
1508
1509
1510
            self.assertLess(n_diff_pixels, 3,
                            "a={}, t={}, s={}, sh={}\n".format(a, t, s, sh) +
                            "n diff pixels={}\n".format(np.sum(np.array(result)[:, :, 0] != true_result[:, :, 0])))
1511
1512
1513

        # Test rotation
        a = 45
ptrblck's avatar
ptrblck committed
1514
        _test_transformation(a=a, t=(0, 0), s=1.0, sh=(0.0, 0.0))
1515
1516
1517

        # Test translation
        t = [10, 15]
ptrblck's avatar
ptrblck committed
1518
        _test_transformation(a=0.0, t=t, s=1.0, sh=(0.0, 0.0))
1519
1520
1521

        # Test scale
        s = 1.2
ptrblck's avatar
ptrblck committed
1522
        _test_transformation(a=0.0, t=(0.0, 0.0), s=s, sh=(0.0, 0.0))
1523
1524

        # Test shear
ptrblck's avatar
ptrblck committed
1525
        sh = [45.0, 25.0]
1526
1527
1528
1529
1530
        _test_transformation(a=0.0, t=(0.0, 0.0), s=1.0, sh=sh)

        # Test rotation, scale, translation, shear
        for a in range(-90, 90, 25):
            for t1 in range(-10, 10, 5):
1531
                for s in [0.75, 0.98, 1.0, 1.2, 1.4]:
1532
                    for sh in range(-15, 15, 5):
ptrblck's avatar
ptrblck committed
1533
                        _test_transformation(a=a, t=(t1, t1), s=s, sh=(sh, sh))
1534

1535
1536
1537
1538
1539
1540
1541
    def test_random_rotation(self):

        with self.assertRaises(ValueError):
            transforms.RandomRotation(-0.7)
            transforms.RandomRotation([-0.7])
            transforms.RandomRotation([-0.7, 0, 0.7])

1542
1543
1544
1545
1546
1547
1548
        # assert fill being either a Sequence or a Number
        with self.assertRaises(TypeError):
            transforms.RandomRotation(0, fill={})

        t = transforms.RandomRotation(0, fill=None)
        self.assertTrue(t.fill == 0)

1549
1550
        t = transforms.RandomRotation(10)
        angle = t.get_params(t.degrees)
1551
        self.assertTrue(angle > -10 and angle < 10)
1552
1553
1554

        t = transforms.RandomRotation((-10, 10))
        angle = t.get_params(t.degrees)
1555
        self.assertTrue(-10 < angle < 10)
1556

1557
1558
1559
        # Checking if RandomRotation can be printed as string
        t.__repr__()

1560
1561
1562
        # assert deprecation warning and non-BC
        with self.assertWarnsRegex(UserWarning, r"Argument resample is deprecated and will be removed"):
            t = transforms.RandomRotation((-10, 10), resample=2)
1563
            self.assertEqual(t.interpolation, transforms.InterpolationMode.BILINEAR)
1564
1565

        # assert changed type warning
1566
        with self.assertWarnsRegex(UserWarning, r"Argument interpolation should be of type InterpolationMode"):
1567
            t = transforms.RandomRotation((-10, 10), interpolation=2)
1568
            self.assertEqual(t.interpolation, transforms.InterpolationMode.BILINEAR)
1569

1570
1571
1572
1573
1574
1575
1576
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
    def test_random_affine(self):

        with self.assertRaises(ValueError):
            transforms.RandomAffine(-0.7)
            transforms.RandomAffine([-0.7])
            transforms.RandomAffine([-0.7, 0, 0.7])

            transforms.RandomAffine([-90, 90], translate=2.0)
            transforms.RandomAffine([-90, 90], translate=[-1.0, 1.0])
            transforms.RandomAffine([-90, 90], translate=[-1.0, 0.0, 1.0])

            transforms.RandomAffine([-90, 90], translate=[0.2, 0.2], scale=[0.0])
            transforms.RandomAffine([-90, 90], translate=[0.2, 0.2], scale=[-1.0, 1.0])
            transforms.RandomAffine([-90, 90], translate=[0.2, 0.2], scale=[0.5, -0.5])
            transforms.RandomAffine([-90, 90], translate=[0.2, 0.2], scale=[0.5, 3.0, -0.5])

            transforms.RandomAffine([-90, 90], translate=[0.2, 0.2], scale=[0.5, 0.5], shear=-7)
            transforms.RandomAffine([-90, 90], translate=[0.2, 0.2], scale=[0.5, 0.5], shear=[-10])
            transforms.RandomAffine([-90, 90], translate=[0.2, 0.2], scale=[0.5, 0.5], shear=[-10, 0, 10])
ptrblck's avatar
ptrblck committed
1589
            transforms.RandomAffine([-90, 90], translate=[0.2, 0.2], scale=[0.5, 0.5], shear=[-10, 0, 10, 0, 10])
1590

1591
1592
1593
1594
1595
1596
1597
        # assert fill being either a Sequence or a Number
        with self.assertRaises(TypeError):
            transforms.RandomAffine(0, fill={})

        t = transforms.RandomAffine(0, fill=None)
        self.assertTrue(t.fill == 0)

1598
1599
1600
        x = np.zeros((100, 100, 3), dtype=np.uint8)
        img = F.to_pil_image(x)

ptrblck's avatar
ptrblck committed
1601
        t = transforms.RandomAffine(10, translate=[0.5, 0.3], scale=[0.7, 1.3], shear=[-10, 10, 20, 40])
1602
1603
1604
        for _ in range(100):
            angle, translations, scale, shear = t.get_params(t.degrees, t.translate, t.scale, t.shear,
                                                             img_size=img.size)
1605
1606
1607
1608
1609
1610
1611
1612
            self.assertTrue(-10 < angle < 10)
            self.assertTrue(-img.size[0] * 0.5 <= translations[0] <= img.size[0] * 0.5,
                            "{} vs {}".format(translations[0], img.size[0] * 0.5))
            self.assertTrue(-img.size[1] * 0.5 <= translations[1] <= img.size[1] * 0.5,
                            "{} vs {}".format(translations[1], img.size[1] * 0.5))
            self.assertTrue(0.7 < scale < 1.3)
            self.assertTrue(-10 < shear[0] < 10)
            self.assertTrue(-20 < shear[1] < 40)
1613
1614
1615
1616

        # Checking if RandomAffine can be printed as string
        t.__repr__()

1617
        t = transforms.RandomAffine(10, interpolation=transforms.InterpolationMode.BILINEAR)
1618
1619
1620
1621
1622
        self.assertIn("bilinear", t.__repr__())

        # assert deprecation warning and non-BC
        with self.assertWarnsRegex(UserWarning, r"Argument resample is deprecated and will be removed"):
            t = transforms.RandomAffine(10, resample=2)
1623
            self.assertEqual(t.interpolation, transforms.InterpolationMode.BILINEAR)
1624
1625
1626
1627
1628
1629

        with self.assertWarnsRegex(UserWarning, r"Argument fillcolor is deprecated and will be removed"):
            t = transforms.RandomAffine(10, fillcolor=10)
            self.assertEqual(t.fill, 10)

        # assert changed type warning
1630
        with self.assertWarnsRegex(UserWarning, r"Argument interpolation should be of type InterpolationMode"):
1631
            t = transforms.RandomAffine(10, interpolation=2)
1632
            self.assertEqual(t.interpolation, transforms.InterpolationMode.BILINEAR)
1633

1634
1635
1636
1637
1638
1639
1640
1641
1642
1643
1644
1645
1646
1647
1648
    def test_to_grayscale(self):
        """Unit tests for grayscale transform"""

        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')
        x_pil_2 = x_pil.convert('L')
        gray_np = np.array(x_pil_2)

        # Test Set: Grayscale an image with desired number of output channels
        # Case 1: RGB -> 1 channel grayscale
        trans1 = transforms.Grayscale(num_output_channels=1)
        gray_pil_1 = trans1(x_pil)
        gray_np_1 = np.array(gray_pil_1)
1649
1650
        self.assertEqual(gray_pil_1.mode, 'L', 'mode should be L')
        self.assertEqual(gray_np_1.shape, tuple(x_shape[0:2]), 'should be 1 channel')
1651
1652
1653
1654
1655
1656
        np.testing.assert_equal(gray_np, gray_np_1)

        # Case 2: RGB -> 3 channel grayscale
        trans2 = transforms.Grayscale(num_output_channels=3)
        gray_pil_2 = trans2(x_pil)
        gray_np_2 = np.array(gray_pil_2)
1657
1658
        self.assertEqual(gray_pil_2.mode, 'RGB', 'mode should be RGB')
        self.assertEqual(gray_np_2.shape, tuple(x_shape), 'should be 3 channel')
1659
1660
1661
1662
1663
1664
1665
1666
        np.testing.assert_equal(gray_np_2[:, :, 0], gray_np_2[:, :, 1])
        np.testing.assert_equal(gray_np_2[:, :, 1], gray_np_2[:, :, 2])
        np.testing.assert_equal(gray_np, gray_np_2[:, :, 0])

        # Case 3: 1 channel grayscale -> 1 channel grayscale
        trans3 = transforms.Grayscale(num_output_channels=1)
        gray_pil_3 = trans3(x_pil_2)
        gray_np_3 = np.array(gray_pil_3)
1667
1668
        self.assertEqual(gray_pil_3.mode, 'L', 'mode should be L')
        self.assertEqual(gray_np_3.shape, tuple(x_shape[0:2]), 'should be 1 channel')
1669
1670
1671
1672
1673
1674
        np.testing.assert_equal(gray_np, gray_np_3)

        # Case 4: 1 channel grayscale -> 3 channel grayscale
        trans4 = transforms.Grayscale(num_output_channels=3)
        gray_pil_4 = trans4(x_pil_2)
        gray_np_4 = np.array(gray_pil_4)
1675
1676
        self.assertEqual(gray_pil_4.mode, 'RGB', 'mode should be RGB')
        self.assertEqual(gray_np_4.shape, tuple(x_shape), 'should be 3 channel')
1677
1678
1679
1680
        np.testing.assert_equal(gray_np_4[:, :, 0], gray_np_4[:, :, 1])
        np.testing.assert_equal(gray_np_4[:, :, 1], gray_np_4[:, :, 2])
        np.testing.assert_equal(gray_np, gray_np_4[:, :, 0])

1681
1682
1683
        # Checking if Grayscale can be printed as string
        trans4.__repr__()

1684
1685
1686
1687
1688
1689
1690
1691
1692
1693
1694
1695
1696
1697
1698
1699
1700
1701
1702
    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_grayscale(self):
        """Unit tests for random grayscale transform"""

        # Test Set 1: RGB -> 3 channel grayscale
        random_state = random.getstate()
        random.seed(42)
        x_shape = [2, 2, 3]
        x_np = np.random.randint(0, 256, x_shape, np.uint8)
        x_pil = Image.fromarray(x_np, mode='RGB')
        x_pil_2 = x_pil.convert('L')
        gray_np = np.array(x_pil_2)

        num_samples = 250
        num_gray = 0
        for _ in range(num_samples):
            gray_pil_2 = transforms.RandomGrayscale(p=0.5)(x_pil)
            gray_np_2 = np.array(gray_pil_2)
            if np.array_equal(gray_np_2[:, :, 0], gray_np_2[:, :, 1]) and \
1703
1704
                    np.array_equal(gray_np_2[:, :, 1], gray_np_2[:, :, 2]) and \
                    np.array_equal(gray_np, gray_np_2[:, :, 0]):
1705
1706
1707
1708
                num_gray = num_gray + 1

        p_value = stats.binom_test(num_gray, num_samples, p=0.5)
        random.setstate(random_state)
1709
        self.assertGreater(p_value, 0.0001)
1710
1711
1712
1713
1714
1715
1716
1717
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729

        # Test Set 2: grayscale -> 1 channel grayscale
        random_state = random.getstate()
        random.seed(42)
        x_shape = [2, 2, 3]
        x_np = np.random.randint(0, 256, x_shape, np.uint8)
        x_pil = Image.fromarray(x_np, mode='RGB')
        x_pil_2 = x_pil.convert('L')
        gray_np = np.array(x_pil_2)

        num_samples = 250
        num_gray = 0
        for _ in range(num_samples):
            gray_pil_3 = transforms.RandomGrayscale(p=0.5)(x_pil_2)
            gray_np_3 = np.array(gray_pil_3)
            if np.array_equal(gray_np, gray_np_3):
                num_gray = num_gray + 1

        p_value = stats.binom_test(num_gray, num_samples, p=1.0)  # Note: grayscale is always unchanged
        random.setstate(random_state)
1730
        self.assertGreater(p_value, 0.0001)
1731
1732
1733
1734
1735
1736
1737
1738
1739
1740
1741
1742
1743

        # Test set 3: Explicit tests
        x_shape = [2, 2, 3]
        x_data = [0, 5, 13, 54, 135, 226, 37, 8, 234, 90, 255, 1]
        x_np = np.array(x_data, dtype=np.uint8).reshape(x_shape)
        x_pil = Image.fromarray(x_np, mode='RGB')
        x_pil_2 = x_pil.convert('L')
        gray_np = np.array(x_pil_2)

        # Case 3a: RGB -> 3 channel grayscale (grayscaled)
        trans2 = transforms.RandomGrayscale(p=1.0)
        gray_pil_2 = trans2(x_pil)
        gray_np_2 = np.array(gray_pil_2)
1744
1745
        self.assertEqual(gray_pil_2.mode, 'RGB', 'mode should be RGB')
        self.assertEqual(gray_np_2.shape, tuple(x_shape), 'should be 3 channel')
1746
1747
1748
1749
1750
1751
1752
1753
        np.testing.assert_equal(gray_np_2[:, :, 0], gray_np_2[:, :, 1])
        np.testing.assert_equal(gray_np_2[:, :, 1], gray_np_2[:, :, 2])
        np.testing.assert_equal(gray_np, gray_np_2[:, :, 0])

        # Case 3b: RGB -> 3 channel grayscale (unchanged)
        trans2 = transforms.RandomGrayscale(p=0.0)
        gray_pil_2 = trans2(x_pil)
        gray_np_2 = np.array(gray_pil_2)
1754
1755
        self.assertEqual(gray_pil_2.mode, 'RGB', 'mode should be RGB')
        self.assertEqual(gray_np_2.shape, tuple(x_shape), 'should be 3 channel')
1756
1757
1758
1759
1760
1761
        np.testing.assert_equal(x_np, gray_np_2)

        # Case 3c: 1 channel grayscale -> 1 channel grayscale (grayscaled)
        trans3 = transforms.RandomGrayscale(p=1.0)
        gray_pil_3 = trans3(x_pil_2)
        gray_np_3 = np.array(gray_pil_3)
1762
1763
        self.assertEqual(gray_pil_3.mode, 'L', 'mode should be L')
        self.assertEqual(gray_np_3.shape, tuple(x_shape[0:2]), 'should be 1 channel')
1764
1765
1766
1767
1768
1769
        np.testing.assert_equal(gray_np, gray_np_3)

        # Case 3d: 1 channel grayscale -> 1 channel grayscale (unchanged)
        trans3 = transforms.RandomGrayscale(p=0.0)
        gray_pil_3 = trans3(x_pil_2)
        gray_np_3 = np.array(gray_pil_3)
1770
1771
        self.assertEqual(gray_pil_3.mode, 'L', 'mode should be L')
        self.assertEqual(gray_np_3.shape, tuple(x_shape[0:2]), 'should be 1 channel')
1772
1773
        np.testing.assert_equal(gray_np, gray_np_3)

1774
1775
1776
        # Checking if RandomGrayscale can be printed as string
        trans3.__repr__()

1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791
1792
1793
1794
1795
1796
1797
1798
1799
1800
1801
1802
1803
1804
1805
1806
1807
1808
1809
1810
1811
1812
1813
1814
1815
1816
1817
1818
    def test_gaussian_blur_asserts(self):
        np_img = np.ones((100, 100, 3), dtype=np.uint8) * 255
        img = F.to_pil_image(np_img, "RGB")

        with self.assertRaisesRegex(ValueError, r"If kernel_size is a sequence its length should be 2"):
            F.gaussian_blur(img, [3])

        with self.assertRaisesRegex(ValueError, r"If kernel_size is a sequence its length should be 2"):
            F.gaussian_blur(img, [3, 3, 3])
        with self.assertRaisesRegex(ValueError, r"Kernel size should be a tuple/list of two integers"):
            transforms.GaussianBlur([3, 3, 3])

        with self.assertRaisesRegex(ValueError, r"kernel_size should have odd and positive integers"):
            F.gaussian_blur(img, [4, 4])
        with self.assertRaisesRegex(ValueError, r"Kernel size value should be an odd and positive number"):
            transforms.GaussianBlur([4, 4])

        with self.assertRaisesRegex(ValueError, r"kernel_size should have odd and positive integers"):
            F.gaussian_blur(img, [-3, -3])
        with self.assertRaisesRegex(ValueError, r"Kernel size value should be an odd and positive number"):
            transforms.GaussianBlur([-3, -3])

        with self.assertRaisesRegex(ValueError, r"If sigma is a sequence, its length should be 2"):
            F.gaussian_blur(img, 3, [1, 1, 1])
        with self.assertRaisesRegex(ValueError, r"sigma should be a single number or a list/tuple with length 2"):
            transforms.GaussianBlur(3, [1, 1, 1])

        with self.assertRaisesRegex(ValueError, r"sigma should have positive values"):
            F.gaussian_blur(img, 3, -1.0)
        with self.assertRaisesRegex(ValueError, r"If sigma is a single number, it must be positive"):
            transforms.GaussianBlur(3, -1.0)

        with self.assertRaisesRegex(TypeError, r"kernel_size should be int or a sequence of integers"):
            F.gaussian_blur(img, "kernel_size_string")
        with self.assertRaisesRegex(ValueError, r"Kernel size should be a tuple/list of two integers"):
            transforms.GaussianBlur("kernel_size_string")

        with self.assertRaisesRegex(TypeError, r"sigma should be either float or sequence of floats"):
            F.gaussian_blur(img, 3, "sigma_string")
        with self.assertRaisesRegex(ValueError, r"sigma should be a single number or a list/tuple with length 2"):
            transforms.GaussianBlur(3, "sigma_string")

1819
1820
1821
1822
1823
1824
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839
1840
1841
1842
1843
1844
1845
1846
1847
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858
1859
1860
1861
1862
1863
1864
1865
1866
1867
1868
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879
1880
1881
1882
1883
1884
1885
1886
1887
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
    def _test_randomness(self, fn, trans, configs):
        random_state = random.getstate()
        random.seed(42)
        img = transforms.ToPILImage()(torch.rand(3, 16, 18))

        for p in [0.5, 0.7]:
            for config in configs:
                inv_img = fn(img, **config)

                num_samples = 250
                counts = 0
                for _ in range(num_samples):
                    tranformation = trans(p=p, **config)
                    tranformation.__repr__()
                    out = tranformation(img)
                    if out == inv_img:
                        counts += 1

                p_value = stats.binom_test(counts, num_samples, p=p)
                random.setstate(random_state)
                self.assertGreater(p_value, 0.0001)

    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_invert(self):
        self._test_randomness(
            F.invert,
            transforms.RandomInvert,
            [{}]
        )

    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_posterize(self):
        self._test_randomness(
            F.posterize,
            transforms.RandomPosterize,
            [{"bits": 4}]
        )

    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_solarize(self):
        self._test_randomness(
            F.solarize,
            transforms.RandomSolarize,
            [{"threshold": 192}]
        )

    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_adjust_sharpness(self):
        self._test_randomness(
            F.adjust_sharpness,
            transforms.RandomAdjustSharpness,
            [{"sharpness_factor": 2.0}]
        )

    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_autocontrast(self):
        self._test_randomness(
            F.autocontrast,
            transforms.RandomAutocontrast,
            [{}]
        )

    @unittest.skipIf(stats is None, 'scipy.stats not available')
    def test_random_equalize(self):
        self._test_randomness(
            F.equalize,
            transforms.RandomEqualize,
            [{}]
        )

    def test_autoaugment(self):
        for policy in transforms.AutoAugmentPolicy:
            for fill in [None, 85, (128, 128, 128)]:
                random.seed(42)
                img = Image.open(GRACE_HOPPER)
                transform = transforms.AutoAugment(policy=policy, fill=fill)
                for _ in range(100):
                    img = transform(img)
                transform.__repr__()

1899

1900
1901
if __name__ == '__main__':
    unittest.main()