File size: 9,711 Bytes
3f75218
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Copyright (c) Facebook, Inc. and its affiliates. All Rights Reserved
"""
Transforms and data augmentation for both image + bbox.
"""
import os
import random

import PIL
import torch
import torchvision.transforms as T
import torchvision.transforms.functional as F

from groundingdino.util.box_ops import box_xyxy_to_cxcywh
from groundingdino.util.misc import interpolate


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", "positive_map"]

    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:
            if field in target:
                target[field] = target[field][keep]

    if os.environ.get("IPDB_SHILONG_DEBUG", None) == "INFO":
        # for debug and visualization only.
        if "strings_positive" in target:
            target["strings_positive"] = [
                _i for _i, _j in zip(target["strings_positive"], keep) if _j
            ]

    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.size[::-1])
    if "masks" in target:
        target["masks"] = torch.nn.functional.pad(target["masks"], (0, padding[0], 0, padding[1]))
    return padded_image, target


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

    def __call__(self, img, target):
        return resize(img, target, self.size)


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 RandomSizeCrop(object):
    def __init__(self, min_size: int, max_size: int, respect_boxes: bool = False):
        # respect_boxes:    True to keep all boxes
        #                   False to tolerence box filter
        self.min_size = min_size
        self.max_size = max_size
        self.respect_boxes = respect_boxes

    def __call__(self, img: PIL.Image.Image, target: dict):
        init_boxes = len(target["boxes"])
        max_patience = 10
        for i in range(max_patience):
            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])
            result_img, result_target = crop(img, target, region)
            if (
                not self.respect_boxes
                or len(result_target["boxes"]) == init_boxes
                or i == max_patience - 1
            ):
                return result_img, result_target
        return result_img, result_target


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) / 2.0))
        crop_left = int(round((image_width - crop_width) / 2.0))
        return crop(img, target, (crop_top, crop_left, crop_height, crop_width))


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


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

    def __call__(self, image, target=None):
        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 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