File size: 21,103 Bytes
483de47
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
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
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
# ------------------------------------------------------------------------
# Copyright (c) 2022 megvii-research. All Rights Reserved.
# ------------------------------------------------------------------------
# Modified from Deformable DETR (https://github.com/fundamentalvision/Deformable-DETR)
# Copyright (c) 2020 SenseTime. All Rights Reserved.
# ------------------------------------------------------------------------
# Modified from DETR (https://github.com/facebookresearch/detr)
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
# ------------------------------------------------------------------------

"""
Transforms and data augmentation for both image + bbox.
"""
import copy
import random
import PIL
import cv2
import torch
import torchvision.transforms as T
import torchvision.transforms.functional as F
from PIL import Image, ImageDraw
from util.box_ops import box_xyxy_to_cxcywh
from util.misc import interpolate
import numpy as np
import os 



def crop_mot(image, target, region):
    cropped_image = F.crop(image, *region)

    target = target.copy()
    i, j, h, w = region

    # should we do something wrt the original size?
    target["size"] = torch.tensor([h, w])

    fields = ["labels", "iscrowd", "obj_ids", "scores"]

    if "boxes" in target:
        boxes = target["boxes"]
        max_size = torch.as_tensor([w, h], dtype=torch.float32)
        cropped_boxes = boxes - torch.as_tensor([j, i, j, i])
        cropped_boxes = torch.min(cropped_boxes.reshape(-1, 2, 2), max_size)
        cropped_boxes = cropped_boxes.clamp(min=0)
        target["boxes"] = cropped_boxes.reshape(-1, 4)
        fields.append("boxes")

    if "masks" in target:
        # FIXME should we update the area here if there are no boxes?
        target['masks'] = target['masks'][:, i:i + h, j:j + w]
        fields.append("masks")

    # remove elements for which the boxes or masks that have zero area
    if "boxes" in target or "masks" in target:
        # favor boxes selection when defining which elements to keep
        # this is compatible with previous implementation
        if "boxes" in target:
            cropped_boxes = target['boxes'].reshape(-1, 2, 2)
            keep = torch.all(cropped_boxes[:, 1, :] > cropped_boxes[:, 0, :], dim=1)
        else:
            keep = target['masks'].flatten(1).any(1)

        for field in fields:
            n_size = len(target[field])
            target[field] = target[field][keep[:n_size]]

    return cropped_image, target


def random_shift(image, target, region, sizes):
    oh, ow = sizes
    # step 1, shift crop and re-scale image firstly
    cropped_image = F.crop(image, *region)
    cropped_image = F.resize(cropped_image, sizes)

    target = target.copy()
    i, j, h, w = region

    # should we do something wrt the original size?
    target["size"] = torch.tensor([h, w])

    fields = ["labels", "scores", "iscrowd", "obj_ids"]

    if "boxes" in target:
        boxes = target["boxes"]
        cropped_boxes = boxes - torch.as_tensor([j, i, j, i])
        cropped_boxes *= torch.as_tensor([ow / w, oh / h, ow / w, oh / h])
        target["boxes"] = cropped_boxes.reshape(-1, 4)
        fields.append("boxes")

    if "masks" in target:
        # FIXME should we update the area here if there are no boxes?
        target['masks'] = target['masks'][:, i:i + h, j:j + w]
        fields.append("masks")

    # remove elements for which the boxes or masks that have zero area
    if "boxes" in target or "masks" in target:
        # favor boxes selection when defining which elements to keep
        # this is compatible with previous implementation
        if "boxes" in target:
            cropped_boxes = target['boxes'].reshape(-1, 2, 2)
            max_size = torch.as_tensor([w, h], dtype=torch.float32)
            cropped_boxes = torch.min(cropped_boxes.reshape(-1, 2, 2), max_size)
            cropped_boxes = cropped_boxes.clamp(min=0)
            keep = torch.all(cropped_boxes[:, 1, :] > cropped_boxes[:, 0, :], dim=1)
        else:
            keep = target['masks'].flatten(1).any(1)

        for field in fields:
            n_size = len(target[field])
            target[field] = target[field][keep[:n_size]]

    return cropped_image, target


def crop(image, target, region):
    cropped_image = F.crop(image, *region)

    target = target.copy()
    i, j, h, w = region

    # should we do something wrt the original size?
    target["size"] = torch.tensor([h, w])

    fields = ["labels", "area", "iscrowd"]
    if 'obj_ids' in target:
        fields.append('obj_ids')

    if "boxes" in target:
        boxes = target["boxes"]
        max_size = torch.as_tensor([w, h], dtype=torch.float32)
        cropped_boxes = boxes - torch.as_tensor([j, i, j, i])
        cropped_boxes = torch.min(cropped_boxes.reshape(-1, 2, 2), max_size)
        cropped_boxes = cropped_boxes.clamp(min=0)

        area = (cropped_boxes[:, 1, :] - cropped_boxes[:, 0, :]).prod(dim=1)
        target["boxes"] = cropped_boxes.reshape(-1, 4)
        target["area"] = area
        fields.append("boxes")

    if "masks" in target:
        # FIXME should we update the area here if there are no boxes?
        target['masks'] = target['masks'][:, i:i + h, j:j + w]
        fields.append("masks")

    # remove elements for which the boxes or masks that have zero area
    if "boxes" in target or "masks" in target:
        # favor boxes selection when defining which elements to keep
        # this is compatible with previous implementation
        if "boxes" in target:
            cropped_boxes = target['boxes'].reshape(-1, 2, 2)
            keep = torch.all(cropped_boxes[:, 1, :] > cropped_boxes[:, 0, :], dim=1)
        else:
            keep = target['masks'].flatten(1).any(1)

        for field in fields:
            target[field] = target[field][keep]

    return cropped_image, target


def hflip(image, target):
    flipped_image = F.hflip(image)

    w, h = image.size

    target = target.copy()
    if "boxes" in target:
        boxes = target["boxes"]
        boxes = boxes[:, [2, 1, 0, 3]] * torch.as_tensor([-1, 1, -1, 1]) + torch.as_tensor([w, 0, w, 0])
        target["boxes"] = boxes

    if "masks" in target:
        target['masks'] = target['masks'].flip(-1)

    return flipped_image, target


def resize(image, target, size, max_size=None):
    # size can be min_size (scalar) or (w, h) tuple

    def get_size_with_aspect_ratio(image_size, size, max_size=None):
        w, h = image_size
        if max_size is not None:
            min_original_size = float(min((w, h)))
            max_original_size = float(max((w, h)))
            if max_original_size / min_original_size * size > max_size:
                size = int(round(max_size * min_original_size / max_original_size))

        if (w <= h and w == size) or (h <= w and h == size):
            return (h, w)

        if w < h:
            ow = size
            oh = int(size * h / w)
        else:
            oh = size
            ow = int(size * w / h)

        return (oh, ow)

    def get_size(image_size, size, max_size=None):
        if isinstance(size, (list, tuple)):
            return size[::-1]
        else:
            return get_size_with_aspect_ratio(image_size, size, max_size)

    size = get_size(image.size, size, max_size)
    rescaled_image = F.resize(image, size)

    if target is None:
        return rescaled_image, None

    ratios = tuple(float(s) / float(s_orig) for s, s_orig in zip(rescaled_image.size, image.size))
    ratio_width, ratio_height = ratios

    target = target.copy()
    if "boxes" in target:
        boxes = target["boxes"]
        scaled_boxes = boxes * torch.as_tensor([ratio_width, ratio_height, ratio_width, ratio_height])
        target["boxes"] = scaled_boxes

    if "area" in target:
        area = target["area"]
        scaled_area = area * (ratio_width * ratio_height)
        target["area"] = scaled_area

    h, w = size
    target["size"] = torch.tensor([h, w])

    if "masks" in target:
        target['masks'] = interpolate(
            target['masks'][:, None].float(), size, mode="nearest")[:, 0] > 0.5

    return rescaled_image, target


def pad(image, target, padding):
    # assumes that we only pad on the bottom right corners
    padded_image = F.pad(image, (0, 0, padding[0], padding[1]))
    if target is None:
        return padded_image, None
    target = target.copy()
    # should we do something wrt the original size?
    target["size"] = torch.tensor(padded_image[::-1])
    if "masks" in target:
        target['masks'] = torch.nn.functional.pad(target['masks'], (0, padding[0], 0, padding[1]))
    return padded_image, target


class MOTHSV:
    def __init__(self, hgain=5, sgain=30, vgain=30) -> None:
        self.hgain = hgain
        self.sgain = sgain
        self.vgain = vgain

    def __call__(self, imgs: list, targets: list):
        hsv_augs = np.random.uniform(-1, 1, 3) * [self.hgain, self.sgain, self.vgain]  # random gains
        hsv_augs *= np.random.randint(0, 2, 3)  # random selection of h, s, v
        hsv_augs = hsv_augs.astype(np.int16)
        for i in range(len(imgs)):
            img = np.array(imgs[i])
            img_hsv = cv2.cvtColor(img, cv2.COLOR_RGB2HSV).astype(np.int16)

            img_hsv[..., 0] = (img_hsv[..., 0] + hsv_augs[0]) % 180
            img_hsv[..., 1] = np.clip(img_hsv[..., 1] + hsv_augs[1], 0, 255)
            img_hsv[..., 2] = np.clip(img_hsv[..., 2] + hsv_augs[2], 0, 255)

            imgs[i] = cv2.cvtColor(img_hsv.astype(img.dtype), cv2.COLOR_HSV2RGB)  # no return needed
        return imgs, targets


class RandomCrop(object):
    def __init__(self, size):
        self.size = size

    def __call__(self, img, target):
        region = T.RandomCrop.get_params(img, self.size)
        return crop(img, target, region)


class MotRandomCrop(RandomCrop):
    def __call__(self, imgs: list, targets: list):
        ret_imgs = []
        ret_targets = []
        region = T.RandomCrop.get_params(imgs[0], self.size)
        for img_i, targets_i in zip(imgs, targets):
            img_i, targets_i = crop(img_i, targets_i, region)
            ret_imgs.append(img_i)
            ret_targets.append(targets_i)
        return ret_imgs, ret_targets

class FixedMotRandomCrop(object):
    def __init__(self, min_size: int, max_size: int):
        self.min_size = min_size
        self.max_size = max_size

    def __call__(self, imgs: list, targets: list):
        ret_imgs = []
        ret_targets = []
        w = random.randint(self.min_size, min(imgs[0].width, self.max_size))
        h = random.randint(self.min_size, min(imgs[0].height, self.max_size))
        region = T.RandomCrop.get_params(imgs[0], [h, w])
        for img_i, targets_i in zip(imgs, targets):
            img_i, targets_i = crop_mot(img_i, targets_i, region)
            ret_imgs.append(img_i)
            ret_targets.append(targets_i)
        return ret_imgs, ret_targets

class MotRandomShift(object):
    def __init__(self, bs=1):
        self.bs = bs

    def __call__(self, imgs: list, targets: list):
        ret_imgs = copy.deepcopy(imgs)
        ret_targets = copy.deepcopy(targets)

        n_frames = len(imgs)
        select_i = random.choice(list(range(n_frames)))
        w, h = imgs[select_i].size

        xshift = (100 * torch.rand(self.bs)).int()
        xshift *= (torch.randn(self.bs) > 0.0).int() * 2 - 1 
        yshift = (100 * torch.rand(self.bs)).int()
        yshift *= (torch.randn(self.bs) > 0.0).int() * 2 - 1
        ymin = max(0, -yshift[0])
        ymax = min(h, h - yshift[0])
        xmin = max(0, -xshift[0])
        xmax = min(w, w - xshift[0])

        region = (int(ymin), int(xmin), int(ymax-ymin), int(xmax-xmin))
        ret_imgs[select_i], ret_targets[select_i] = random_shift(imgs[select_i], targets[select_i], region, (h,w)) 
        
        return ret_imgs, ret_targets


class FixedMotRandomShift(object):
    def __init__(self, bs=1, padding=50):
        self.bs = bs
        self.padding = padding

    def __call__(self, imgs: list, targets: list):
        ret_imgs = []
        ret_targets = []

        n_frames = self.bs
        w, h = imgs[0].size
        xshift = (self.padding * torch.rand(self.bs)).int() + 1
        xshift *= (torch.randn(self.bs) > 0.0).int() * 2 - 1
        yshift = (self.padding * torch.rand(self.bs)).int() + 1
        yshift *= (torch.randn(self.bs) > 0.0).int() * 2 - 1
        ret_imgs.append(imgs[0])
        ret_targets.append(targets[0])
        for i in range(1, n_frames):
            ymin = max(0, -yshift[0])
            ymax = min(h, h - yshift[0])
            xmin = max(0, -xshift[0])
            xmax = min(w, w - xshift[0])
            prev_img = ret_imgs[i-1].copy()
            prev_target = copy.deepcopy(ret_targets[i-1])
            region = (int(ymin), int(xmin), int(ymax - ymin), int(xmax - xmin))
            img_i, target_i = random_shift(prev_img, prev_target, region, (h, w))
            ret_imgs.append(img_i)
            ret_targets.append(target_i)

        return ret_imgs, ret_targets


class RandomSizeCrop(object):
    def __init__(self, min_size: int, max_size: int):
        self.min_size = min_size
        self.max_size = max_size

    def __call__(self, img: PIL.Image.Image, target: dict):
        w = random.randint(self.min_size, min(img.width, self.max_size))
        h = random.randint(self.min_size, min(img.height, self.max_size))
        region = T.RandomCrop.get_params(img, [h, w])
        return crop(img, target, region)


class MotRandomSizeCrop(RandomSizeCrop):
    def __call__(self, imgs, targets):
        w = random.randint(self.min_size, min(imgs[0].width, self.max_size))
        h = random.randint(self.min_size, min(imgs[0].height, self.max_size))
        region = T.RandomCrop.get_params(imgs[0], [h, w])
        ret_imgs = []
        ret_targets = []
        for img_i, targets_i in zip(imgs, targets):
            img_i, targets_i = crop(img_i, targets_i, region)
            ret_imgs.append(img_i)
            ret_targets.append(targets_i)
        return ret_imgs, ret_targets


class CenterCrop(object):
    def __init__(self, size):
        self.size = size

    def __call__(self, img, target):
        image_width, image_height = img.size
        crop_height, crop_width = self.size
        crop_top = int(round((image_height - crop_height) / 2.))
        crop_left = int(round((image_width - crop_width) / 2.))
        return crop(img, target, (crop_top, crop_left, crop_height, crop_width))


class MotCenterCrop(CenterCrop):
    def __call__(self, imgs, targets):
        image_width, image_height = imgs[0].size
        crop_height, crop_width = self.size
        crop_top = int(round((image_height - crop_height) / 2.))
        crop_left = int(round((image_width - crop_width) / 2.))
        ret_imgs = []
        ret_targets = []
        for img_i, targets_i in zip(imgs, targets):
            img_i, targets_i = crop(img_i, targets_i, (crop_top, crop_left, crop_height, crop_width))
            ret_imgs.append(img_i)
            ret_targets.append(targets_i)
        return ret_imgs, ret_targets


class RandomHorizontalFlip(object):
    def __init__(self, p=0.5):
        self.p = p

    def __call__(self, img, target):
        if random.random() < self.p:
            return hflip(img, target)
        return img, target


class MotRandomHorizontalFlip(RandomHorizontalFlip):
    def __call__(self, imgs, targets):
        if random.random() < self.p:
            ret_imgs = []
            ret_targets = []
            for img_i, targets_i in zip(imgs, targets):
                img_i, targets_i = hflip(img_i, targets_i)
                ret_imgs.append(img_i)
                ret_targets.append(targets_i)
            return ret_imgs, ret_targets
        return imgs, targets


class RandomResize(object):
    def __init__(self, sizes, max_size=None):
        assert isinstance(sizes, (list, tuple))
        self.sizes = sizes
        self.max_size = max_size

    def __call__(self, img, target=None):
        size = random.choice(self.sizes)
        return resize(img, target, size, self.max_size)


class MotRandomResize(RandomResize):
    def __call__(self, imgs, targets):
        size = random.choice(self.sizes)
        ret_imgs = []
        ret_targets = []
        for img_i, targets_i in zip(imgs, targets):
            img_i, targets_i = resize(img_i, targets_i, size, self.max_size)
            ret_imgs.append(img_i)
            ret_targets.append(targets_i)
        return ret_imgs, ret_targets


class RandomPad(object):
    def __init__(self, max_pad):
        self.max_pad = max_pad

    def __call__(self, img, target):
        pad_x = random.randint(0, self.max_pad)
        pad_y = random.randint(0, self.max_pad)
        return pad(img, target, (pad_x, pad_y))


class MotRandomPad(RandomPad):
    def __call__(self, imgs, targets):
        pad_x = random.randint(0, self.max_pad)
        pad_y = random.randint(0, self.max_pad)
        ret_imgs = []
        ret_targets = []
        for img_i, targets_i in zip(imgs, targets):
            img_i, target_i = pad(img_i, targets_i, (pad_x, pad_y))
            ret_imgs.append(img_i)
            ret_targets.append(targets_i)
        return ret_imgs, ret_targets


class RandomSelect(object):
    """
    Randomly selects between transforms1 and transforms2,
    with probability p for transforms1 and (1 - p) for transforms2
    """
    def __init__(self, transforms1, transforms2, p=0.5):
        self.transforms1 = transforms1
        self.transforms2 = transforms2
        self.p = p

    def __call__(self, img, target):
        if random.random() < self.p:
            return self.transforms1(img, target)
        return self.transforms2(img, target)


class MotRandomSelect(RandomSelect):
    """
    Randomly selects between transforms1 and transforms2,
    with probability p for transforms1 and (1 - p) for transforms2
    """
    def __call__(self, imgs, targets):
        if random.random() < self.p:
            return self.transforms1(imgs, targets)
        return self.transforms2(imgs, targets)


class ToTensor(object):
    def __call__(self, img, target):
        return F.to_tensor(img), target


class MotToTensor(ToTensor):
    def __call__(self, imgs, targets):
        ret_imgs = []
        for img in imgs:
            ret_imgs.append(F.to_tensor(img))
        return ret_imgs, targets


class RandomErasing(object):

    def __init__(self, *args, **kwargs):
        self.eraser = T.RandomErasing(*args, **kwargs)

    def __call__(self, img, target):
        return self.eraser(img), target


class MotRandomErasing(RandomErasing):
    def __call__(self, imgs, targets):
        # TODO: Rewrite this part to ensure the data augmentation is same to each image.
        ret_imgs = []
        for img_i, targets_i in zip(imgs, targets):
            ret_imgs.append(self.eraser(img_i))
        return ret_imgs, targets


class MoTColorJitter(T.ColorJitter):
    def __call__(self, imgs, targets):
        transform = self.get_params(self.brightness, self.contrast,
                                    self.saturation, self.hue)
        ret_imgs = []
        for img_i, targets_i in zip(imgs, targets):
            ret_imgs.append(transform(img_i))
        return ret_imgs, targets


class Normalize(object):
    def __init__(self, mean, std):
        self.mean = mean
        self.std = std

    def __call__(self, image, target=None):
        if target is not None:
            target['ori_img'] = image.clone()
        image = F.normalize(image, mean=self.mean, std=self.std)
        if target is None:
            return image, None
        target = target.copy()
        h, w = image.shape[-2:]
        if "boxes" in target:
            boxes = target["boxes"]
            boxes = box_xyxy_to_cxcywh(boxes)
            boxes = boxes / torch.tensor([w, h, w, h], dtype=torch.float32)
            target["boxes"] = boxes
        return image, target


class MotNormalize(Normalize):
    def __call__(self, imgs, targets=None):
        ret_imgs = []
        ret_targets = []
        for i in range(len(imgs)):
            img_i = imgs[i]
            targets_i = targets[i] if targets is not None else None
            img_i, targets_i = super().__call__(img_i, targets_i)
            ret_imgs.append(img_i)
            ret_targets.append(targets_i)
        return ret_imgs, ret_targets


class Compose(object):
    def __init__(self, transforms):
        self.transforms = transforms

    def __call__(self, image, target):
        for t in self.transforms:
            image, target = t(image, target)
        return image, target

    def __repr__(self):
        format_string = self.__class__.__name__ + "("
        for t in self.transforms:
            format_string += "\n"
            format_string += "    {0}".format(t)
        format_string += "\n)"
        return format_string


class MotCompose(Compose):
    def __call__(self, imgs, targets):
        for t in self.transforms:
            imgs, targets = t(imgs, targets)
        return imgs, targets