Skip to content

augment_transform

data.transform.augmentation.augment_transform

Implement ImageTransform classes for image augmentation.

ImageType module-attribute 

ImageType = ndarray

ParamsType module-attribute 

ParamsType = Dict[str, Any]

MaskType module-attribute 

MaskType = ndarray

BBoxesType module-attribute 

BBoxesType = Union[ndarray, NumpyBoxes]

PolygonType module-attribute 

PolygonType = List[ndarray]

OffsetType module-attribute 

OffsetType = Tuple[float, float]

BaseTransform 

BaseTransform(prob: float = 0.5)

Transform의 기본 interface를 정의합니다.

Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
14
15
16
def __init__(self, prob: float = 0.5) -> None:
    self.prob = prob
    self._additional_targets = {}

prob instance-attribute 

prob = prob

_additional_targets instance-attribute 

_additional_targets = {}

data_for_params property 

data_for_params: List[str]

targets property 

targets: Dict[str, Callable]

__call__ 

__call__(*args, force_apply: bool = False, **data) -> Any
Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
def __call__(self, *args, force_apply: bool = False, **data) -> Any:
    if args:
        raise KeyError("Data must be passed as named arguments, for example: aug(image=image)")

    if (random.random() < self.prob) or force_apply:
        params = {}
        if self.data_for_params:
            data_for_params = {k: data[k] for k in self.data_for_params}
            params_from_data = self.get_params_from_data(data_for_params)
            params.update(params_from_data)

        apply_param = self.get_apply_params()
        params.update(apply_param)
        data = self.apply_with_params(params, **data)

    return data

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
35
36
def get_apply_params(self) -> ParamsType:
    return {}

get_params_from_data 

get_params_from_data(data_for_params: ParamsType) -> ParamsType

이 함수는 input으로부터 parameter들을 뽑을 때 사용됩니다.

Parameters:

  • data_for_params (ParamsType) –

    params을 추출할 데이터를 입력으로 갖습니다.

Returns:

  • ParamsType ( ParamsType ) –

    params로 쓰일 데이터를 반환합니다.

Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
42
43
44
45
46
47
48
49
50
51
def get_params_from_data(self, data_for_params: ParamsType) -> ParamsType:
    """이 함수는 input으로부터 parameter들을 뽑을 때 사용됩니다.

    Args:
        data_for_params (ParamsType): params을 추출할 데이터를 입력으로 갖습니다.

    Returns:
        ParamsType: params로 쓰일 데이터를 반환합니다.
    """
    return {}

apply_with_params 

apply_with_params(params: ParamsType, **data) -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
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
def apply_with_params(self, params: ParamsType, **data) -> ParamsType:
    def _apply_with_params(method: callable, target: str, data: Dict, params: ParamsType):
        output = []

        if isinstance(data, List) and target == "image":
            is_multipage = True
        else:
            is_multipage = False
            data = [data]

        for d in data:
            output.append(method(d, **params))

        if not is_multipage:
            output = output[0]

        return output

    for target, target_method in self.targets.items():
        if target in data:
            data[target] = _apply_with_params(target_method, target, data[target], params)
    for additional_target, target in self._additional_targets.items():
        if additional_target in data:
            target_method = self.targets[target]
            data[additional_target] = _apply_with_params(
                target_method, target, data[additional_target], params
            )
    return data

add_targets 

add_targets(additional_targets: Dict[str, str]) -> None
Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
86
87
def add_targets(self, additional_targets: Dict[str, str]) -> None:
    self._additional_targets = additional_targets

ImageTransform 

ImageTransform(prob: float = 0.5)

Bases: BaseTransform

Image와 라벨에 적용되는 Transform 입니다.

Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
14
15
16
def __init__(self, prob: float = 0.5) -> None:
    self.prob = prob
    self._additional_targets = {}

targets property 

targets: Dict[str, Callable]

apply_to_image 

apply_to_image(image: ImageType, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
48
49
def apply_to_image(self, image: ImageType, **params) -> ImageType:
    return image

apply_to_mask 

apply_to_mask(mask: MaskType, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
51
52
def apply_to_mask(self, mask: MaskType, **params) -> MaskType:
    return mask

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
54
55
def apply_to_bboxes(self, bboxes: BBoxesType, **params) -> BBoxesType:
    return bboxes

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
57
58
def apply_to_polygons(self, polygons: PolygonType, **params) -> PolygonType:
    return polygons

ImageSizeParams

data_for_params property 

data_for_params: List[str]

get_params_from_data 

get_params_from_data(data_for_params: ParamsType) -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
75
76
77
78
79
80
def get_params_from_data(self, data_for_params: ParamsType) -> ParamsType:
    img = data_for_params["image"]
    if not isinstance(img, list):
        return {"image_size": read_image_size(img)}
    else:
        return {"image_size": read_image_size(img[0])}

ToNumpy 

ToNumpy(prob: float = 0.5)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
14
15
16
def __init__(self, prob: float = 0.5) -> None:
    self.prob = prob
    self._additional_targets = {}

apply_to_image 

apply_to_image(image: ImageType, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
84
85
def apply_to_image(self, image: ImageType, **params) -> ImageType:
    return to_numpy(image=image)

apply_to_mask 

apply_to_mask(mask: MaskType, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
87
88
def apply_to_mask(self, mask: MaskType, **params) -> MaskType:
    return to_numpy(image=mask)

ToPil 

ToPil(prob: float = 0.5)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
14
15
16
def __init__(self, prob: float = 0.5) -> None:
    self.prob = prob
    self._additional_targets = {}

apply_to_image 

apply_to_image(image: ImageType, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
92
93
def apply_to_image(self, image: ImageType, **params) -> ImageType:
    return to_pil(image=image)

apply_to_mask 

apply_to_mask(mask: MaskType, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
95
96
def apply_to_mask(self, mask: MaskType, **params) -> MaskType:
    return to_pil(image=mask)

VerticalFlip 

VerticalFlip(prob: float = 0.5)

Bases: ImageSizeParams, ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
14
15
16
def __init__(self, prob: float = 0.5) -> None:
    self.prob = prob
    self._additional_targets = {}

apply_to_image 

apply_to_image(image: ImageType, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
100
101
def apply_to_image(self, image: ImageType, **params) -> ImageType:
    return vertical_flip(image=image)

apply_to_mask 

apply_to_mask(mask: MaskType, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
103
104
def apply_to_mask(self, mask: MaskType, **params) -> MaskType:
    return vertical_flip(image=mask)

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, image_size: Tuple[int], **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
106
107
def apply_to_bboxes(self, bboxes: BBoxesType, image_size: Tuple[int], **params) -> BBoxesType:
    return box_function.vflip_box(bboxes=bboxes, image_size=image_size)

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, image_size: Tuple[int], **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
109
110
def apply_to_polygons(self, polygons: PolygonType, image_size: Tuple[int], **params) -> PolygonType:
    return polygon_function.vertical_flip(polygons=polygons, image_size=image_size)

HorizontalFlip 

HorizontalFlip(prob: float = 0.5)

Bases: ImageSizeParams, ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/base_transform.py 
14
15
16
def __init__(self, prob: float = 0.5) -> None:
    self.prob = prob
    self._additional_targets = {}

apply_to_image 

apply_to_image(image: ImageType, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
114
115
def apply_to_image(self, image: ImageType, **params) -> ImageType:
    return horizontal_flip(image=image)

apply_to_mask 

apply_to_mask(mask: MaskType, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
117
118
def apply_to_mask(self, mask: MaskType, **params) -> MaskType:
    return horizontal_flip(image=mask)

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, image_size: Tuple[int], **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
120
121
def apply_to_bboxes(self, bboxes: BBoxesType, image_size: Tuple[int], **params) -> BBoxesType:
    return box_function.hflip_box(bboxes=bboxes, image_size=image_size)

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, image_size: Tuple[int], **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
123
124
def apply_to_polygons(self, polygons: PolygonType, image_size: Tuple[int], **params) -> PolygonType:
    return polygon_function.horizontal_flip(polygons=polygons, image_size=image_size)

Rotate 

Rotate(angle_limit: Optional[List[float]] = None, interpolation: int = cv2.INTER_LINEAR, border_mode: int = cv2.BORDER_CONSTANT, value: Union[int, float, List[int], List[float]] = 0, mask_value: Union[int, float] = 0, crop_border: bool = False, **kwargs)

Bases: ImageSizeParams, ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
def __init__(
    self,
    angle_limit: Optional[List[float]] = None,
    interpolation: int = cv2.INTER_LINEAR,
    border_mode: int = cv2.BORDER_CONSTANT,
    value: Union[int, float, List[int], List[float]] = 0,
    mask_value: Union[int, float] = 0,
    crop_border: bool = False,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if angle_limit is None:
        angle_limit = [-30.0, 30.0]

    check_range(angle_limit, -360.0, 360.0)

    self.angle_limit = angle_limit
    self.interpolation = interpolation
    self.border_mode = border_mode
    self.value = value
    self.mask_value = mask_value
    self.crop_border = crop_border

angle_limit instance-attribute 

angle_limit = angle_limit

interpolation instance-attribute 

interpolation = interpolation

border_mode instance-attribute 

border_mode = border_mode

value instance-attribute 

value = value

mask_value instance-attribute 

mask_value = mask_value

crop_border instance-attribute 

crop_border = crop_border

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
152
153
def get_apply_params(self) -> ParamsType:
    return {"angle": random.uniform(*self.angle_limit)}

apply_to_image 

apply_to_image(image: ImageType, angle: float = 0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
155
156
157
158
159
160
161
162
163
def apply_to_image(self, image: ImageType, angle: float = 0, **params) -> ImageType:
    return rotate(
        image=image,
        angle=angle,
        interpolation=self.interpolation,
        border_mode=self.border_mode,
        value=self.value,
        crop_border=self.crop_border,
    )

apply_to_mask 

apply_to_mask(mask: MaskType, angle: float = 0, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
165
166
167
168
169
170
171
172
173
def apply_to_mask(self, mask: MaskType, angle: float = 0, **params) -> MaskType:
    return rotate(
        image=mask,
        angle=angle,
        interpolation=cv2.INTER_NEAREST,
        border_mode=self.border_mode,
        value=self.mask_value,
        crop_border=self.crop_border,
    )

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, image_size: Tuple[int], angle: float = 0, **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
175
176
177
178
179
180
181
182
def apply_to_bboxes(
    self,
    bboxes: BBoxesType,
    image_size: Tuple[int],
    angle: float = 0,
    **params,
) -> BBoxesType:
    return box_function.rotate(bboxes=bboxes, angle=angle, image_size=image_size)

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, angle: float, image_size: Tuple[int], **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
184
185
186
187
188
189
190
191
def apply_to_polygons(
    self,
    polygons: PolygonType,
    angle: float,
    image_size: Tuple[int],
    **params,
) -> PolygonType:
    return polygon_function.rotate(polygons=polygons, angle=angle, image_size=image_size)

RandomRotate90 

RandomRotate90(**kwargs)

Bases: ImageSizeParams, ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
195
196
def __init__(self, **kwargs) -> None:
    super().__init__(**kwargs)

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
198
199
def get_apply_params(self) -> ParamsType:
    return {"factor": random.randint(0, 3)}

apply_to_image 

apply_to_image(image: ImageType, factor: int = 0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
201
202
203
204
205
def apply_to_image(self, image: ImageType, factor: int = 0, **params) -> ImageType:
    return random_rotate90(
        image=image,
        factor=factor,
    )

apply_to_mask 

apply_to_mask(mask: MaskType, factor: int = 0, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
207
208
209
210
211
def apply_to_mask(self, mask: MaskType, factor: int = 0, **params) -> MaskType:
    return random_rotate90(
        image=mask,
        factor=factor,
    )

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, image_size: Tuple[int], factor: int = 0, **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
213
214
215
216
217
218
219
220
def apply_to_bboxes(
    self,
    bboxes: BBoxesType,
    image_size: Tuple[int],
    factor: int = 0,
    **params,
) -> BBoxesType:
    return box_function.rotate90(bboxes=bboxes, factor=factor, image_size=image_size)

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, image_size: Tuple[int], factor: int = 0, **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
222
223
224
225
226
227
228
229
def apply_to_polygons(
    self,
    polygons: PolygonType,
    image_size: Tuple[int],
    factor: int = 0,
    **params,
) -> PolygonType:
    return polygon_function.rotate90(polygons=polygons, factor=factor, image_size=image_size)

ColorJitter 

ColorJitter(brightness_limit: Optional[List[float]] = None, contrast_limit: Optional[List[float]] = None, saturation_limit: Optional[List[float]] = None, hue_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
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
def __init__(
    self,
    brightness_limit: Optional[List[float]] = None,
    contrast_limit: Optional[List[float]] = None,
    saturation_limit: Optional[List[float]] = None,
    hue_limit: Optional[List[float]] = None,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if brightness_limit is None:
        brightness_limit = [0.8, 1.2]
    if contrast_limit is None:
        contrast_limit = [0.8, 1.2]
    if saturation_limit is None:
        saturation_limit = [0.8, 1.2]
    if hue_limit is None:
        hue_limit = [-0.2, 0.2]

    check_range(brightness_limit, 0.01, 10.00)
    check_range(contrast_limit, 0.01, 10.00)
    check_range(saturation_limit, 0.01, 10.00)
    check_range(hue_limit, -0.50, 0.50)

    self.brightness_limit = brightness_limit
    self.contrast_limit = contrast_limit
    self.saturation_limit = saturation_limit
    self.hue_limit = hue_limit

brightness_limit instance-attribute 

brightness_limit = brightness_limit

contrast_limit instance-attribute 

contrast_limit = contrast_limit

saturation_limit instance-attribute 

saturation_limit = saturation_limit

hue_limit instance-attribute 

hue_limit = hue_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
262
263
264
265
266
267
268
269
270
271
272
def get_apply_params(self) -> ParamsType:
    order = [0, 1, 2, 3]
    random.shuffle(order)

    return {
        "brightness": random.uniform(*self.brightness_limit),
        "contrast": random.uniform(*self.contrast_limit),
        "saturation": random.uniform(*self.saturation_limit),
        "hue": random.uniform(*self.hue_limit),
        "order": order,
    }

apply_to_image 

apply_to_image(image: ImageType, brightness: float = 1.0, contrast: float = 1.0, saturation: float = 1.0, hue: float = 0, order: List[int] = [0, 1, 2, 3], **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
def apply_to_image(
    self,
    image: ImageType,
    brightness: float = 1.0,
    contrast: float = 1.0,
    saturation: float = 1.0,
    hue: float = 0,
    order: List[int] = [0, 1, 2, 3],
    **params,
) -> ImageType:
    return color_jitter(
        image=image,
        brightness=brightness,
        contrast=contrast,
        saturation=saturation,
        hue=hue,
        order=order,
    )

Blur 

Blur(ksize_limit: Optional[List[int]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
295
296
297
298
299
300
301
302
303
def __init__(self, ksize_limit: Optional[List[int]] = None, **kwargs) -> None:
    super().__init__(**kwargs)

    if ksize_limit is None:
        ksize_limit = [3, 7]

    check_range(ksize_limit, 1, 100)

    self.ksize_limit = ksize_limit

ksize_limit instance-attribute 

ksize_limit = ksize_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
305
306
def get_apply_params(self) -> ParamsType:
    return {"ksize": random.randint(*self.ksize_limit)}

apply_to_image 

apply_to_image(image: ImageType, ksize: int = 3, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
308
309
def apply_to_image(self, image: ImageType, ksize: int = 3, **params) -> ImageType:
    return blur(image=image, ksize=ksize)

GaussianBlur 

GaussianBlur(ksize_limit: Optional[List[int]] = None, sigma_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
def __init__(
    self,
    ksize_limit: Optional[List[int]] = None,
    sigma_limit: Optional[List[float]] = None,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if ksize_limit is None:
        ksize_limit = [3, 7]
    if sigma_limit is None:
        sigma_limit = [0.0, 0.0]

    check_range(ksize_limit, 1, 100)
    check_range(sigma_limit, 0.00, 100.00)

    self.ksize_limit = ksize_limit
    self.sigma_limit = sigma_limit

ksize_limit instance-attribute 

ksize_limit = ksize_limit

sigma_limit instance-attribute 

sigma_limit = sigma_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
332
333
334
335
336
def get_apply_params(self) -> ParamsType:
    return {
        "ksize": random.randint(*self.ksize_limit),
        "sigma": random.uniform(*self.sigma_limit),
    }

apply_to_image 

apply_to_image(image: ImageType, ksize: int = 3, sigma: float = 0.0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
338
339
340
341
def apply_to_image(
    self, image: ImageType, ksize: int = 3, sigma: float = 0.0, **params
) -> ImageType:
    return gaussian_blur(image=image, ksize=ksize, sigma=sigma)

AdjustBrightness 

AdjustBrightness(brightness_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
345
346
347
348
349
350
351
352
353
def __init__(self, brightness_limit: Optional[List[float]] = None, **kwargs) -> None:
    super().__init__(**kwargs)

    if brightness_limit is None:
        brightness_limit = [-0.2, 0.2]

    check_range(brightness_limit, -1.0, 1.0)

    self.brightness_limit = brightness_limit

brightness_limit instance-attribute 

brightness_limit = brightness_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
355
356
def get_apply_params(self) -> ParamsType:
    return {"brightness": random.uniform(*self.brightness_limit)}

apply_to_image 

apply_to_image(image: ImageType, brightness: float = 0.0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
358
359
def apply_to_image(self, image: ImageType, brightness: float = 0.0, **params) -> ImageType:
    return adjust_brightness(image=image, brightness=brightness)

AdjustContrast 

AdjustContrast(contrast_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
363
364
365
366
367
368
369
370
371
def __init__(self, contrast_limit: Optional[List[float]] = None, **kwargs) -> None:
    super().__init__(**kwargs)

    if contrast_limit is None:
        contrast_limit = [-0.2, 0.2]

    check_range(contrast_limit, -1.0, 1.0)

    self.contrast_limit = contrast_limit

contrast_limit instance-attribute 

contrast_limit = contrast_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
373
374
def get_apply_params(self) -> ParamsType:
    return {"contrast": random.uniform(*self.contrast_limit)}

apply_to_image 

apply_to_image(image: ImageType, contrast: float = 0.0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
376
377
def apply_to_image(self, image: ImageType, contrast: float = 0.0, **params) -> ImageType:
    return adjust_contrast(image=image, contrast=contrast)

AdjustHue 

AdjustHue(hue_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
381
382
383
384
385
386
387
388
389
def __init__(self, hue_limit: Optional[List[float]] = None, **kwargs) -> None:
    super().__init__(**kwargs)

    if hue_limit is None:
        hue_limit = [-0.2, 0.2]

    check_range(hue_limit, -1.0, 1.0)

    self.hue_limit = hue_limit

hue_limit instance-attribute 

hue_limit = hue_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
391
392
def get_apply_params(self) -> ParamsType:
    return {"hue": random.uniform(*self.hue_limit)}

apply_to_image 

apply_to_image(image: ImageType, hue: float = 0.0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
394
395
def apply_to_image(self, image: ImageType, hue: float = 0.0, **params) -> ImageType:
    return adjust_hue(image=image, hue=hue)

AdjustSaturation 

AdjustSaturation(saturation_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
399
400
401
402
403
404
405
406
407
def __init__(self, saturation_limit: Optional[List[float]] = None, **kwargs) -> None:
    super().__init__(**kwargs)

    if saturation_limit is None:
        saturation_limit = [-0.2, 0.2]

    check_range(saturation_limit, -1.0, 1.0)

    self.saturation_limit = saturation_limit

saturation_limit instance-attribute 

saturation_limit = saturation_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
409
410
def get_apply_params(self) -> ParamsType:
    return {"saturation": random.uniform(*self.saturation_limit)}

apply_to_image 

apply_to_image(image: ImageType, saturation: float = 0.0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
412
413
def apply_to_image(self, image: ImageType, saturation: float = 0.0, **params) -> ImageType:
    return adjust_saturation(image=image, saturation=saturation)

AdjustGamma 

AdjustGamma(gamma_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
417
418
419
420
421
422
423
424
425
def __init__(self, gamma_limit: Optional[List[float]] = None, **kwargs) -> None:
    super().__init__(**kwargs)

    if gamma_limit is None:
        gamma_limit = [-0.2, 0.2]

    check_range(gamma_limit, -1.0, 1.0)

    self.gamma_limit = gamma_limit

gamma_limit instance-attribute 

gamma_limit = gamma_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
427
428
def get_apply_params(self) -> ParamsType:
    return {"gamma": random.uniform(*self.gamma_limit)}

apply_to_image 

apply_to_image(image: ImageType, gamma: float = 0.0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
430
431
def apply_to_image(self, image: ImageType, gamma: float = 0.0, **params) -> ImageType:
    return adjust_gamma(image=image, gamma=gamma)

AdjustBrightnessContrast 

AdjustBrightnessContrast(brightness_limit: Optional[List[float]] = None, contrast_limit: Optional[List[float]] = None, brightness_by_max: bool = True, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
def __init__(
    self,
    brightness_limit: Optional[List[float]] = None,
    contrast_limit: Optional[List[float]] = None,
    brightness_by_max: bool = True,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if brightness_limit is None:
        brightness_limit = [-0.2, 0.2]
    if contrast_limit is None:
        contrast_limit = [-0.2, 0.2]

    check_range(brightness_limit, -1.00, 1.00)
    check_range(contrast_limit, -1.00, 1.00)

    self.brightness_limit = brightness_limit
    self.contrast_limit = contrast_limit
    self.brightness_by_max = brightness_by_max

brightness_limit instance-attribute 

brightness_limit = brightness_limit

contrast_limit instance-attribute 

contrast_limit = contrast_limit

brightness_by_max instance-attribute 

brightness_by_max = brightness_by_max

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
456
457
458
459
460
def get_apply_params(self) -> ParamsType:
    return {
        "brightness": random.uniform(*self.brightness_limit),
        "contrast": random.uniform(*self.contrast_limit),
    }

apply_to_image 

apply_to_image(image: ImageType, brightness: float = 0.0, contrast: float = 0.0, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
462
463
464
465
466
467
468
469
470
def apply_to_image(
    self, image: ImageType, brightness: float = 0.0, contrast: float = 0.0, **params
) -> ImageType:
    return adjust_brightness_contrast(
        image=image,
        brightness=brightness,
        contrast=contrast,
        brightness_by_max=self.brightness_by_max,
    )

IsoNoise 

IsoNoise(color_shift_limit: Optional[List[float]] = None, intensity_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
def __init__(
    self,
    color_shift_limit: Optional[List[float]] = None,
    intensity_limit: Optional[List[float]] = None,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if color_shift_limit is None:
        color_shift_limit = [0.01, 0.05]
    if intensity_limit is None:
        intensity_limit = [0.1, 0.5]

    check_range(color_shift_limit, 0.00, 1.00)
    check_range(intensity_limit, 0.00, 2.00)

    self.color_shift_limit = color_shift_limit
    self.intensity_limit = intensity_limit

color_shift_limit instance-attribute 

color_shift_limit = color_shift_limit

intensity_limit instance-attribute 

intensity_limit = intensity_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
493
494
495
496
497
498
def get_apply_params(self) -> ParamsType:
    return {
        "color_shift": random.uniform(*self.color_shift_limit),
        "intensity": random.uniform(*self.intensity_limit),
        "random_state": random.randint(0, 65536),
    }

apply_to_image 

apply_to_image(image: ImageType, color_shift: float = 0.05, intensity: float = 0.5, random_state: Optional[int] = None, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
500
501
502
503
504
505
506
507
508
509
510
def apply_to_image(
    self,
    image: ImageType,
    color_shift: float = 0.05,
    intensity: float = 0.50,
    random_state: Optional[int] = None,
    **params,
) -> ImageType:
    return iso_noise(
        image=image, color_shift=color_shift, intensity=intensity, random_state=random_state
    )

JpegCompression 

JpegCompression(quality_limit: Optional[List[int]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
514
515
516
517
518
519
520
521
522
523
524
525
526
def __init__(
    self,
    quality_limit: Optional[List[int]] = None,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if quality_limit is None:
        quality_limit = [5, 30]

    check_range(quality_limit, 0, 100)

    self.quality_limit = quality_limit

quality_limit instance-attribute 

quality_limit = quality_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
528
529
def get_apply_params(self) -> ParamsType:
    return {"quality": random.randint(*self.quality_limit)}

apply_to_image 

apply_to_image(image: ImageType, quality: int = 100) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
531
532
533
534
535
536
def apply_to_image(
    self,
    image: ImageType,
    quality: int = 100,
) -> ImageType:
    return image_compression(image=image, quality=quality, image_type=".jpeg")

Sharpen 

Sharpen(alpha_limit: Optional[List[int]] = None, lightness_limit: Optional[List[int]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
def __init__(
    self,
    alpha_limit: Optional[List[int]] = None,
    lightness_limit: Optional[List[int]] = None,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if alpha_limit is None:
        alpha_limit = [0.20, 0.50]

    if lightness_limit is None:
        lightness_limit = [0.80, 1.20]

    check_range(alpha_limit, 0.00, 1.00)
    check_range(lightness_limit, 0.00, 10.00)

    self.alpha_limit = alpha_limit
    self.lightness_limit = lightness_limit

alpha_limit instance-attribute 

alpha_limit = alpha_limit

lightness_limit instance-attribute 

lightness_limit = lightness_limit

_generate_sharpening_matrix 

_generate_sharpening_matrix(alpha: float, lightness: float) -> ndarray
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
560
561
562
563
564
565
566
567
def _generate_sharpening_matrix(self, alpha: float, lightness: float) -> np.ndarray:
    matrix_nochange = np.array([[0, 0, 0], [0, 1, 0], [0, 0, 0]], dtype=np.float32)
    matrix_effect = np.array(
        [[-1, -1, -1], [-1, 8 + lightness, -1], [-1, -1, -1]],
        dtype=np.float32,
    )
    matrix = (1 - alpha) * matrix_nochange + alpha * matrix_effect
    return matrix

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
569
570
571
572
573
def get_apply_params(self) -> ParamsType:
    sharpening_matrix = self._generate_sharpening_matrix(
        random.uniform(*self.alpha_limit), random.uniform(*self.lightness_limit)
    )
    return {"sharpening_matrix": sharpening_matrix}

apply_to_image 

apply_to_image(image: ImageType, sharpening_matrix: ndarray) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
575
576
577
578
579
580
def apply_to_image(
    self,
    image: ImageType,
    sharpening_matrix: np.ndarray,
) -> ImageType:
    return sharpen(image=image, sharpening_matrix=sharpening_matrix)

MultiplicativeNoise 

MultiplicativeNoise(multiplier: Optional[List[float]] = None, per_channel: bool = True, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
def __init__(
    self,
    multiplier: Optional[List[float]] = None,
    per_channel: bool = True,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if multiplier is None:
        multiplier = [0.80, 1.20]

    check_range(multiplier, 0.00, 5.00)

    self.multiplier = multiplier
    self.per_channel = per_channel

multiplier instance-attribute 

multiplier = multiplier

per_channel instance-attribute 

per_channel = per_channel

_generate_multiplier 

_generate_multiplier(image: ImageType, multiplier: ndarray, random_seed: int) -> ndarray
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
600
601
602
603
604
605
606
607
608
609
610
611
612
def _generate_multiplier(
    self, image: ImageType, multiplier: np.ndarray, random_seed: int
) -> np.ndarray:
    if multiplier[0] == multiplier[1]:
        return np.array([multiplier[0]])

    height, width = image.shape[:2]
    num_channels = 3

    result_multiplier = np.random.RandomState(seed=random_seed).uniform(
        multiplier[0], multiplier[1], [height, width, num_channels]
    )
    return result_multiplier

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
614
615
616
617
618
619
620
621
622
623
def get_apply_params(self) -> ParamsType:
    def _shrink_range(low: int, high: int) -> List[int]:
        lower_bound = random.uniform(low, high)
        upper_bound = random.uniform(lower_bound, high)
        return [lower_bound, upper_bound]

    return {
        "multiplier": np.array(_shrink_range(*self.multiplier)),
        "random_seed": random.randint(0, 65536),
    }

apply_to_image 

apply_to_image(image: ImageType, multiplier: ndarray, random_seed: int) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
625
626
627
def apply_to_image(self, image: ImageType, multiplier: np.ndarray, random_seed: int) -> ImageType:
    _multiplier = self._generate_multiplier(image, multiplier, random_seed)
    return multiplicative_noise(image=image, multiplier=_multiplier)

RatioJitter 

RatioJitter(proportion_limit: Optional[Union[List[float], int, float]] = None, resampling: str = 'bilinear', border_mode: int = cv2.BORDER_CONSTANT, value: Union[int, float, List[int], List[float]] = 0, mask_value: Union[int, float] = 0, **kwargs)

Bases: ImageSizeParams, ImageTransform

crop, pad, and resize to original image size

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
def __init__(
    self,
    proportion_limit: Optional[Union[List[float], int, float]] = None,
    resampling: str = "bilinear",
    border_mode: int = cv2.BORDER_CONSTANT,
    value: Union[int, float, List[int], List[float]] = 0,
    mask_value: Union[int, float] = 0,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if proportion_limit is None:
        proportion_limit = [-0.10, 0.10]

    if isinstance(proportion_limit, int):
        # NOTE: int 입력은 백분율로 적용.
        proportion_limit = [-abs(proportion_limit) / 100, abs(proportion_limit) / 100]

    if isinstance(proportion_limit, float):
        # NOTE: float 입력은 0~1 스케일로 적용.
        proportion_limit = [-abs(proportion_limit), abs(proportion_limit)]

    check_range(proportion_limit, -0.50, 0.50)

    self.proportion_limit = proportion_limit
    self.resampling = resampling
    self.border_mode = border_mode
    self.value = value
    self.mask_value = mask_value

proportion_limit instance-attribute 

proportion_limit = proportion_limit

resampling instance-attribute 

resampling = resampling

border_mode instance-attribute 

border_mode = border_mode

value instance-attribute 

value = value

mask_value instance-attribute 

mask_value = mask_value

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
663
664
665
666
667
668
669
def get_apply_params(self) -> ParamsType:
    return {
        "proportion_left": random.uniform(*self.proportion_limit),
        "proportion_right": random.uniform(*self.proportion_limit),
        "proportion_top": random.uniform(*self.proportion_limit),
        "proportion_bottom": random.uniform(*self.proportion_limit),
    }

apply_to_image 

apply_to_image(image: ImageType, proportion_left: float, proportion_right: float, proportion_top: float, proportion_bottom: float, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
def apply_to_image(
    self,
    image: ImageType,
    proportion_left: float,
    proportion_right: float,
    proportion_top: float,
    proportion_bottom: float,
    **params,
) -> ImageType:
    return ratio_jitter(
        image=image,
        proportion_left=proportion_left,
        proportion_right=proportion_right,
        proportion_top=proportion_top,
        proportion_bottom=proportion_bottom,
        resampling=self.resampling,
        border_mode=self.border_mode,
        value=self.value,
    )

apply_to_mask 

apply_to_mask(mask: MaskType, proportion_left: float, proportion_right: float, proportion_top: float, proportion_bottom: float, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
def apply_to_mask(
    self,
    mask: MaskType,
    proportion_left: float,
    proportion_right: float,
    proportion_top: float,
    proportion_bottom: float,
    **params,
) -> MaskType:
    return ratio_jitter(
        image=mask,
        proportion_left=proportion_left,
        proportion_right=proportion_right,
        proportion_top=proportion_top,
        proportion_bottom=proportion_bottom,
        resampling="nearest",
        border_mode=self.border_mode,
        value=self.mask_value,
    )

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, image_size: Tuple[int], proportion_left: float, proportion_right: float, proportion_top: float, proportion_bottom: float, **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
def apply_to_bboxes(
    self,
    bboxes: BBoxesType,
    image_size: Tuple[int],
    proportion_left: float,
    proportion_right: float,
    proportion_top: float,
    proportion_bottom: float,
    **params,
) -> BBoxesType:
    return box_function.ratio_jitter(
        bboxes=bboxes,
        image_size=image_size,
        proportion_left=proportion_left,
        proportion_right=proportion_right,
        proportion_top=proportion_top,
        proportion_bottom=proportion_bottom,
    )

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, image_size: Tuple[int], proportion_left: float, proportion_right: float, proportion_top: float, proportion_bottom: float, **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
def apply_to_polygons(
    self,
    polygons: PolygonType,
    image_size: Tuple[int],
    proportion_left: float,
    proportion_right: float,
    proportion_top: float,
    proportion_bottom: float,
    **params,
) -> PolygonType:
    return polygon_function.ratio_jitter(
        polygons=polygons,
        image_size=image_size,
        proportion_left=proportion_left,
        proportion_right=proportion_right,
        proportion_top=proportion_top,
        proportion_bottom=proportion_bottom,
    )

Zoom 

Zoom(ratio_limit: Optional[List[float]] = None, resampling: str = 'bilinear', border_mode: int = cv2.BORDER_CONSTANT, value: Union[int, float, List[int], List[float]] = 0, mask_value: Union[int, float] = 0, **kwargs)

Bases: ImageSizeParams, ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
def __init__(
    self,
    ratio_limit: Optional[List[float]] = None,
    resampling: str = "bilinear",
    border_mode: int = cv2.BORDER_CONSTANT,
    value: Union[int, float, List[int], List[float]] = 0,
    mask_value: Union[int, float] = 0,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if ratio_limit is None:
        ratio_limit = [0.50, 2.00]

    check_range(ratio_limit, 0.01, 100.00)

    self.ratio_limit = ratio_limit
    self.resampling = resampling
    self.border_mode = border_mode
    self.value = value
    self.mask_value = mask_value

ratio_limit instance-attribute 

ratio_limit = ratio_limit

resampling instance-attribute 

resampling = resampling

border_mode instance-attribute 

border_mode = border_mode

value instance-attribute 

value = value

mask_value instance-attribute 

mask_value = mask_value

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
773
774
775
776
777
778
779
def get_apply_params(self) -> ParamsType:
    log_ratio = (math.log(self.ratio_limit[0]), math.log(self.ratio_limit[1]))
    return {
        "ratio": math.exp(random.uniform(*log_ratio)),
        "h_start": random.random(),
        "w_start": random.random(),
    }

apply_to_image 

apply_to_image(image: ImageType, ratio: float, h_start: float, w_start: float, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
def apply_to_image(
    self,
    image: ImageType,
    ratio: float,
    h_start: float,
    w_start: float,
    **params,
) -> ImageType:
    return zoom(
        image=image,
        ratio=ratio,
        h_start=h_start,
        w_start=w_start,
        resampling=self.resampling,
        border_mode=self.border_mode,
        value=self.value,
    )

apply_to_mask 

apply_to_mask(mask: MaskType, ratio: float, h_start: float, w_start: float, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
def apply_to_mask(
    self,
    mask: MaskType,
    ratio: float,
    h_start: float,
    w_start: float,
    **params,
) -> MaskType:
    return zoom(
        image=mask,
        ratio=ratio,
        h_start=h_start,
        w_start=w_start,
        resampling="nearest",
        border_mode=self.border_mode,
        value=self.mask_value,
    )

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, image_size: Tuple[int], ratio: float, h_start: float, w_start: float, **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
def apply_to_bboxes(
    self,
    bboxes: BBoxesType,
    image_size: Tuple[int],
    ratio: float,
    h_start: float,
    w_start: float,
    **params,
) -> BBoxesType:
    return box_function.zoom(
        bboxes=bboxes,
        image_size=image_size,
        ratio=ratio,
        h_start=h_start,
        w_start=w_start,
    )

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, image_size: Tuple[int], ratio: float, h_start: float, w_start: float, **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
def apply_to_polygons(
    self,
    polygons: PolygonType,
    image_size: Tuple[int],
    ratio: float,
    h_start: float,
    w_start: float,
    **params,
) -> PolygonType:
    return polygon_function.zoom(
        polygons=polygons,
        image_size=image_size,
        ratio=ratio,
        h_start=h_start,
        w_start=w_start,
    )

RandomResizedCrop 

RandomResizedCrop(scale_limit: Optional[List[float]] = None, aspect_ratio_limit: Optional[List[float]] = None, resampling: str = 'bilinear', **kwargs)

Bases: ImageSizeParams, ImageTransform

Crop a random part of the input and rescale it to original size

Parameters:

  • scale_limit (Optional[List[float]], default: None ) –

    range of size of the origin size cropped. Defaults to [0.45, 1.00].

  • aspect_ratio_limit (Optional[List[float]], default: None ) –

    range of aspect ratio of the origin aspect ratio cropped. Defaults to [0.50, 2.00].

  • resampling (str, default: 'bilinear' ) –

    interpolation method. Defaults to "bilinear".

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
def __init__(
    self,
    scale_limit: Optional[List[float]] = None,
    aspect_ratio_limit: Optional[List[float]] = None,
    resampling: str = "bilinear",
    **kwargs,
) -> None:
    """
    Args:
        scale_limit (Optional[List[float]], optional): range of size of the origin size cropped. Defaults to [0.45, 1.00].
        aspect_ratio_limit (Optional[List[float]], optional): range of aspect ratio of the origin aspect ratio cropped. Defaults to [0.50, 2.00].
        resampling (str, optional): interpolation method. Defaults to "bilinear".
    """
    super().__init__(**kwargs)

    if scale_limit is None:
        scale_limit = [0.45, 1.00]
    if aspect_ratio_limit is None:
        aspect_ratio_limit = [0.50, 2.00]

    check_range(scale_limit, 0.01, 1.00)
    check_range(aspect_ratio_limit, 0.10, 10.00)

    self.scale_limit = scale_limit
    self.aspect_ratio_limit = aspect_ratio_limit
    self.resampling = resampling

scale_limit instance-attribute 

scale_limit = scale_limit

aspect_ratio_limit instance-attribute 

aspect_ratio_limit = aspect_ratio_limit

resampling instance-attribute 

resampling = resampling

targets property 

targets: Dict[str, Callable]

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
def get_apply_params(self) -> ParamsType:
    log_aspect_ratio = (math.log(self.aspect_ratio_limit[0]), math.log(self.aspect_ratio_limit[1]))
    w_scale, h_scale = 1.0, 1.0
    num_attempts = 10
    for _ in range(num_attempts):
        scale = random.uniform(*self.scale_limit)
        aspect_ratio = math.exp(random.uniform(*log_aspect_ratio))
        _w_scale = scale * math.sqrt(aspect_ratio)
        _h_scale = scale / math.sqrt(aspect_ratio)
        if _w_scale <= 1.0 and _h_scale <= 1.0:
            w_scale, h_scale = _w_scale, _h_scale
            break

    return {
        "w_scale": w_scale,
        "h_scale": h_scale,
        "h_start": random.random(),
        "w_start": random.random(),
    }

apply_to_image 

apply_to_image(image: ImageType, h_scale: float, w_scale: float, h_start: float, w_start: float, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
def apply_to_image(
    self,
    image: ImageType,
    h_scale: float,
    w_scale: float,
    h_start: float,
    w_start: float,
    **params,
) -> ImageType:
    return random_resized_crop(
        image=image,
        h_scale=h_scale,
        w_scale=w_scale,
        h_start=h_start,
        w_start=w_start,
        resampling=self.resampling,
    )

apply_to_mask 

apply_to_mask(mask: MaskType, h_scale: float, w_scale: float, h_start: float, w_start: float, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
def apply_to_mask(
    self,
    mask: MaskType,
    h_scale: float,
    w_scale: float,
    h_start: float,
    w_start: float,
    **params,
) -> MaskType:
    return random_resized_crop(
        image=mask,
        h_scale=h_scale,
        w_scale=w_scale,
        h_start=h_start,
        w_start=w_start,
        resampling="nearest",
    )

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
938
939
def apply_to_bboxes(self, bboxes: BBoxesType, **params) -> BBoxesType:
    raise NotImplementedError

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
941
942
def apply_to_polygons(self, polygons: PolygonType, **params) -> PolygonType:
    raise NotImplementedError

RandomResizedCropAndPad 

RandomResizedCropAndPad(scale_limit: Optional[List[float]] = None, aspect_ratio_limit: Optional[List[float]] = None, height: Optional[int] = None, width: Optional[int] = None, resampling: str = 'bilinear', border_mode: int = cv2.BORDER_CONSTANT, value: Union[int, float, List[int], List[float]] = 0, mask_value: Union[int, float] = 0, **kwargs)

Bases: ImageSizeParams, ImageTransform

pad, crop and resize to original image size

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
def __init__(
    self,
    scale_limit: Optional[List[float]] = None,
    aspect_ratio_limit: Optional[List[float]] = None,
    height: Optional[int] = None,
    width: Optional[int] = None,
    resampling: str = "bilinear",
    border_mode: int = cv2.BORDER_CONSTANT,
    value: Union[int, float, List[int], List[float]] = 0,
    mask_value: Union[int, float] = 0,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if scale_limit is None:
        scale_limit = [0.45, 1.00]
    if aspect_ratio_limit is None:
        aspect_ratio_limit = [0.50, 2.00]

    check_range(scale_limit, 0.01, 1.00)
    check_range(aspect_ratio_limit, 0.10, 10.00)

    self.height = height
    self.width = width
    self.scale_limit = scale_limit
    self.aspect_ratio_limit = aspect_ratio_limit
    self.resampling = resampling
    self.border_mode = border_mode
    self.value = value
    self.mask_value = mask_value

height instance-attribute 

height = height

width instance-attribute 

width = width

scale_limit instance-attribute 

scale_limit = scale_limit

aspect_ratio_limit instance-attribute 

aspect_ratio_limit = aspect_ratio_limit

resampling instance-attribute 

resampling = resampling

border_mode instance-attribute 

border_mode = border_mode

value instance-attribute 

value = value

mask_value instance-attribute 

mask_value = mask_value

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
986
987
988
989
990
991
992
993
def get_apply_params(self) -> ParamsType:
    log_aspect_ratio = (math.log(self.aspect_ratio_limit[0]), math.log(self.aspect_ratio_limit[1]))
    return {
        "scale": random.uniform(*self.scale_limit),
        "aspect_ratio": math.exp(random.uniform(*log_aspect_ratio)),
        "h_start": random.random(),
        "w_start": random.random(),
    }

apply_to_image 

apply_to_image(image: ImageType, scale: float, aspect_ratio: float, h_start: float, w_start: float, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
 995
 996
 997
 998
 999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
def apply_to_image(
    self,
    image: ImageType,
    scale: float,
    aspect_ratio: float,
    h_start: float,
    w_start: float,
    **params,
) -> ImageType:
    return random_resized_crop_and_pad(
        image=image,
        scale=scale,
        aspect_ratio=aspect_ratio,
        h_start=h_start,
        w_start=w_start,
        height=self.height,
        width=self.width,
        resampling=self.resampling,
        border_mode=self.border_mode,
        value=self.value,
    )

apply_to_mask 

apply_to_mask(mask: MaskType, scale: float, aspect_ratio: float, h_start: float, w_start: float, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
1028
1029
1030
1031
1032
1033
1034
1035
1036
1037
def apply_to_mask(
    self,
    mask: MaskType,
    scale: float,
    aspect_ratio: float,
    h_start: float,
    w_start: float,
    **params,
) -> MaskType:
    return random_resized_crop_and_pad(
        image=mask,
        scale=scale,
        aspect_ratio=aspect_ratio,
        h_start=h_start,
        w_start=w_start,
        height=self.height,
        width=self.width,
        resampling="nearest",
        border_mode=self.border_mode,
        value=self.mask_value,
    )

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, image_size: Tuple[int], scale: float, aspect_ratio: float, h_start: float, w_start: float, **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1039
1040
1041
1042
1043
1044
1045
1046
1047
1048
1049
1050
1051
1052
1053
1054
1055
1056
1057
1058
def apply_to_bboxes(
    self,
    bboxes: BBoxesType,
    image_size: Tuple[int],
    scale: float,
    aspect_ratio: float,
    h_start: float,
    w_start: float,
    **params,
) -> BBoxesType:
    return box_function.random_resized_crop_and_pad(
        bboxes=bboxes,
        image_size=image_size,
        scale=scale,
        aspect_ratio=aspect_ratio,
        h_start=h_start,
        w_start=w_start,
        height=self.height,
        width=self.width,
    )

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, image_size: Tuple[int], scale: float, aspect_ratio: float, h_start: float, w_start: float, **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1060
1061
1062
1063
1064
1065
1066
1067
1068
1069
1070
1071
1072
1073
1074
1075
1076
1077
def apply_to_polygons(
    self,
    polygons: PolygonType,
    image_size: Tuple[int],
    scale: float,
    aspect_ratio: float,
    h_start: float,
    w_start: float,
    **params,
) -> PolygonType:
    return polygon_function.random_resized_crop_and_pad(
        polygons=polygons,
        image_size=image_size,
        scale=scale,
        aspect_ratio=aspect_ratio,
        h_start=h_start,
        w_start=w_start,
    )

LightReflect 

LightReflect(radius_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1081
1082
1083
1084
1085
1086
1087
1088
1089
def __init__(self, radius_limit: Optional[List[float]] = None, **kwargs) -> None:
    super().__init__(**kwargs)

    if radius_limit is None:
        radius_limit = [0.10, 0.50]

    check_range(radius_limit, 0.00, 1.00)

    self.radius_limit = radius_limit

radius_limit instance-attribute 

radius_limit = radius_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1091
1092
1093
1094
1095
1096
1097
1098
def get_apply_params(self) -> ParamsType:
    return {
        "xc": random.random(),
        "yc": random.random(),
        "x_radius": random.uniform(*self.radius_limit),
        "y_radius": random.uniform(*self.radius_limit),
        "angle": random.uniform(0.00, 360.00),
    }

apply_to_image 

apply_to_image(image: ImageType, xc: float, yc: float, x_radius: float, y_radius: float, angle: float, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1100
1101
1102
1103
1104
1105
1106
1107
1108
1109
1110
1111
1112
def apply_to_image(
    self,
    image: ImageType,
    xc: float,
    yc: float,
    x_radius: float,
    y_radius: float,
    angle: float,
    **params,
) -> ImageType:
    return light_reflect(
        image=image, xc=xc, yc=yc, x_radius=x_radius, y_radius=y_radius, angle=angle
    )

PerspectiveTransform 

PerspectiveTransform(intensity_limit: Optional[List[float]] = None, **kwargs)

Bases: ImageSizeParams, ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
def __init__(
    self,
    intensity_limit: Optional[List[float]] = None,
    **kwargs,
) -> None:
    super().__init__(**kwargs)

    if intensity_limit is None:
        intensity_limit = [0, 30]

    check_range(intensity_limit, 0, 49)

    self.intensity_limit = intensity_limit

intensity_limit instance-attribute 

intensity_limit = intensity_limit

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1130
1131
1132
1133
1134
1135
1136
1137
def get_apply_params(self) -> ParamsType:
    intensity = random.uniform(*self.intensity_limit)
    return {
        "offset_top_left": tuple((random.uniform(0, intensity), random.uniform(0, intensity))),
        "offset_top_right": tuple((random.uniform(0, intensity), random.uniform(0, intensity))),
        "offset_bottom_right": tuple((random.uniform(0, intensity), random.uniform(0, intensity))),
        "offset_bottom_left": tuple((random.uniform(0, intensity), random.uniform(0, intensity))),
    }

apply_to_image 

apply_to_image(image: ImageType, offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1139
1140
1141
1142
1143
1144
1145
1146
1147
1148
1149
1150
1151
1152
1153
1154
1155
def apply_to_image(
    self,
    image: ImageType,
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
    **params,
) -> ImageType:
    return perspective_transform(
        image=image,
        offset_top_left=offset_top_left,
        offset_top_right=offset_top_right,
        offset_bottom_right=offset_bottom_right,
        offset_bottom_left=offset_bottom_left,
        interpolate_method="linear",
    )

apply_to_mask 

apply_to_mask(mask: MaskType, offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType, **params) -> MaskType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1157
1158
1159
1160
1161
1162
1163
1164
1165
1166
1167
1168
1169
1170
1171
1172
1173
def apply_to_mask(
    self,
    mask: MaskType,
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
    **params,
) -> MaskType:
    return perspective_transform(
        image=mask,
        offset_top_left=offset_top_left,
        offset_top_right=offset_top_right,
        offset_bottom_right=offset_bottom_right,
        offset_bottom_left=offset_bottom_left,
        interpolate_method="nearest",
    )

apply_to_bboxes 

apply_to_bboxes(bboxes: BBoxesType, image_size: Tuple[int], offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType, **params) -> BBoxesType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1175
1176
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
def apply_to_bboxes(
    self,
    bboxes: BBoxesType,
    image_size: Tuple[int],
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
    **params,
) -> BBoxesType:
    return box_function.perspective_transform(
        bboxes=bboxes,
        image_size=image_size,
        offset_top_left=offset_top_left,
        offset_top_right=offset_top_right,
        offset_bottom_right=offset_bottom_right,
        offset_bottom_left=offset_bottom_left,
    )

apply_to_polygons 

apply_to_polygons(polygons: PolygonType, image_size: Tuple[int], offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType, **params) -> PolygonType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1194
1195
1196
1197
1198
1199
1200
1201
1202
1203
1204
1205
1206
1207
1208
1209
1210
1211
def apply_to_polygons(
    self,
    polygons: PolygonType,
    image_size: Tuple[int],
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
    **params,
) -> PolygonType:
    return polygon_function.perspective_transform(
        polygons=polygons,
        image_size=image_size,
        offset_top_left=offset_top_left,
        offset_top_right=offset_top_right,
        offset_bottom_right=offset_bottom_right,
        offset_bottom_left=offset_bottom_left,
    )

RandomErasing 

RandomErasing(scale: Tuple[float, float] = (0.02, 0.33), ratio: Tuple[float, float] = (0.2, 3.3), value: Union[int, Tuple[int, int, int], str] = 0, randomly_select_values: bool = False, **kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1215
1216
1217
1218
1219
1220
1221
1222
1223
1224
1225
1226
1227
def __init__(
    self,
    scale: Tuple[float, float] = (0.02, 0.33),
    ratio: Tuple[float, float] = (0.2, 3.3),
    value: Union[int, Tuple[int, int, int], str] = 0,
    randomly_select_values: bool = False,
    **kwargs,
) -> None:
    super().__init__(**kwargs)
    self.scale = scale
    self.ratio = ratio
    self._value = value
    self.randomly_select_values = randomly_select_values

scale instance-attribute 

scale = scale

ratio instance-attribute 

ratio = ratio

_value instance-attribute 

_value = value

randomly_select_values instance-attribute 

randomly_select_values = randomly_select_values

_get_random_erase_params 

_get_random_erase_params()

This is modified version of get_params of random erasing see: https://pytorch.org/vision/main/_modules/torchvision/transforms/transforms.html#RandomErasing.forward

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1229
1230
1231
1232
1233
1234
1235
1236
1237
1238
1239
1240
1241
1242
1243
1244
1245
1246
1247
1248
1249
1250
1251
1252
1253
def _get_random_erase_params(self):
    """
    This is modified version of get_params of random erasing
    see: https://pytorch.org/vision/main/_modules/torchvision/transforms/transforms.html#RandomErasing.forward
    """
    area = 1

    log_ratio = torch.log(torch.tensor(self.ratio))
    for _ in range(10):
        erase_area = area * torch.empty(1).uniform_(self.scale[0], self.scale[1]).item()
        aspect_ratio = torch.exp(torch.empty(1).uniform_(log_ratio[0], log_ratio[1])).item()

        h_in_ratio = math.sqrt(erase_area * aspect_ratio)
        w_in_ratio = math.sqrt(erase_area / aspect_ratio)

        if not (h_in_ratio < 1 and w_in_ratio < 1):
            continue

        x_in_ratio = random.uniform(0, 1 - w_in_ratio)
        y_in_ratio = random.uniform(0, 1 - h_in_ratio)

        return x_in_ratio, y_in_ratio, h_in_ratio, w_in_ratio

    # Return Original Image
    return 0, 0, 1.0, 1.0

get_apply_params 

get_apply_params() -> ParamsType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1255
1256
1257
1258
1259
1260
1261
1262
1263
1264
1265
1266
1267
1268
1269
1270
1271
1272
1273
def get_apply_params(self) -> ParamsType:
    value = self._value
    if self.randomly_select_values:
        value_options = {
            "erase": 0,
            "rgb": (random.randint(0, 255), random.randint(0, 255), random.randint(0, 255)),
            "random": "random",
        }
        value = value_options[random.choice(list(value_options.keys()))]

    x_in_ratio, y_in_ratio, h_in_ratio, w_in_ratio = self._get_random_erase_params()

    return dict(
        x_in_ratio=x_in_ratio,
        y_in_ratio=y_in_ratio,
        h_in_ratio=h_in_ratio,
        w_in_ratio=w_in_ratio,
        value=value,
    )

apply_to_image 

apply_to_image(image: ImageType, x_in_ratio: float, y_in_ratio: float, h_in_ratio: float, w_in_ratio: float, value: Union[int, Tuple[int, int, int], str], **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1275
1276
1277
1278
1279
1280
1281
1282
1283
1284
1285
1286
1287
1288
1289
1290
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
def apply_to_image(
    self,
    image: ImageType,
    x_in_ratio: float,
    y_in_ratio: float,
    h_in_ratio: float,
    w_in_ratio: float,
    value: Union[int, Tuple[int, int, int], str],
    **params,
) -> ImageType:
    is_gray = image.ndim == 2
    if is_gray:
        img_height, img_width = image.shape
    else:
        img_height, img_width, _ = image.shape

    x = int(min(x_in_ratio * img_width, img_width))
    y = int(min(y_in_ratio * img_height, img_height))
    h = int(min(h_in_ratio * img_height, img_height))
    w = int(min(w_in_ratio * img_width, img_width))

    # cast value to script acceptable type
    if isinstance(value, (int, float)):
        value = np.array([float(value)]).astype("uint8")[None, None, :]
    elif isinstance(value, str):
        value = (np.random.normal(size=(h, w, 3)) * 255).astype("uint8")
    elif isinstance(value, (list, tuple)):
        value = np.array([float(v) for v in value]).astype("uint8")[None, None, :]

    return erase(image=image, x=x, y=y, w=w, h=h, v=value)

Grayscale 

Grayscale(**kwargs)

Bases: ImageTransform

Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1308
1309
def __init__(self, **kwargs) -> None:
    super().__init__(**kwargs)

apply_to_image 

apply_to_image(image: ImageType, **params) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_transform.py 
1311
1312
def apply_to_image(self, image: ImageType, **params) -> ImageType:
    return grayscale(image=image)

to_numpy 

to_numpy(image: Union[Image, Tensor, ndarray], copy: bool = False) -> ndarray
Source code in SaigeToolkit/data/transform/image_function.py 
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
def to_numpy(image: Union[Image.Image, torch.Tensor, np.ndarray], copy: bool = False) -> np.ndarray:
    if isinstance(image, Image.Image):
        image = np.array(image)  # Image.Image -> np.ndarray
    elif isinstance(image, torch.Tensor):
        assert image.ndim in [3, 4]  # BCHW / CHW

        if image.ndim == 4:  # if BCHW
            assert image.shape[0] == 1  # Batch size must be 1
            image = image.squeeze(0)  # BCHW -> CHW

        image = image.permute(1, 2, 0)  # CHW -> HWC

        if image.shape[2] == 1:  # if gray image
            image = image.squeeze(2)  # HWC -> HW

        image = image.detach().cpu().numpy()  # torch.Tensor, HW(C) -> np.ndarray, HW(C)
    elif isinstance(image, np.ndarray):
        if copy:
            image = image.copy()
    else:
        raise NotImplementedError

    # gray: HW / rgb & rgba: HWC
    return image

to_pil 

to_pil(image: Union[Image, Tensor, ndarray], copy: bool = False) -> Image
Source code in SaigeToolkit/data/transform/image_function.py 
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
def to_pil(image: Union[Image.Image, torch.Tensor, np.ndarray], copy: bool = False) -> Image.Image:
    if isinstance(image, Image.Image):
        if copy:
            image = image.copy()
    elif isinstance(image, torch.Tensor):
        image = to_numpy(image)  # torch.Tensor -> np.ndarray
        image = Image.fromarray(image)  # np.ndarray -> Image.Image
    elif isinstance(image, np.ndarray):
        assert image.ndim in [2, 3]  # HW / HWC

        if image.ndim == 3 and image.shape[2] == 1:  # grayscale shape HW1
            image = np.squeeze(image, axis=2)  # HW1 -> HW

        image = Image.fromarray(image)  # np.ndarray -> Image.Image
    else:
        raise NotImplementedError

    return image

read_image_size 

read_image_size(image: Union[Image, Tensor, ndarray]) -> ImageSizeType

image size: (W, H)

Source code in SaigeToolkit/data/transform/image_function.py 
11
12
13
14
15
16
17
18
19
20
def read_image_size(image: Union[Image.Image, torch.Tensor, np.ndarray]) -> ImageSizeType:
    """image size: (W, H)"""
    if isinstance(image, Image.Image):
        return image.size
    elif isinstance(image, torch.Tensor) and image.ndim in (2, 3, 4):  # HW, CHW, BCHW
        return (image.shape[-1], image.shape[-2])
    elif isinstance(image, np.ndarray) and image.ndim in (2, 3):  # HW, HWC
        return (image.shape[1], image.shape[0])
    else:
        raise NotImplementedError

vertical_flip 

vertical_flip(image: Union[ImageType, MaskType]) -> Union[ImageType, MaskType]
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
42
43
def vertical_flip(image: Union[ImageType, MaskType]) -> Union[ImageType, MaskType]:
    return AF.vflip(image)

horizontal_flip 

horizontal_flip(image: Union[ImageType, MaskType]) -> Union[ImageType, MaskType]
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
46
47
48
49
50
51
52
def horizontal_flip(image: Union[ImageType, MaskType]) -> Union[ImageType, MaskType]:
    if image.ndim == 3 and image.shape[2] > 1 and image.dtype == np.uint8:
        # Opencv is faster than numpy only in case of
        # non-gray scale 8bits images
        return AF.hflip_cv2(image)

    return AF.hflip(image)

rotate 

rotate(image: Union[ImageType, MaskType], angle: float = 0, interpolation: int = cv2.INTER_LINEAR, border_mode: int = cv2.BORDER_REFLECT_101, value: Union[int, float, List[int], List[float]] = 0, crop_border: bool = False) -> Union[ImageType, MaskType]
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
 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
def rotate(
    image: Union[ImageType, MaskType],
    angle: float = 0,
    interpolation: int = cv2.INTER_LINEAR,
    border_mode: int = cv2.BORDER_REFLECT_101,
    value: Union[int, float, List[int], List[float]] = 0,
    crop_border: bool = False,
) -> Union[ImageType, MaskType]:
    def _rotated_rect_with_max_area(h, w, angle):
        """
        Given a rectangle of size wxh that has been rotated by 'angle' (in
        degrees), computes the width and height of the largest possible
        axis-aligned rectangle (maximal area) within the rotated rectangle.

        Code from: https://stackoverflow.com/questions/16702966/rotate-image-and-crop-out-black-borders
        """

        angle = math.radians(angle)
        width_is_longer = w >= h
        side_long, side_short = (w, h) if width_is_longer else (h, w)

        # since the solutions for angle, -angle and 180-angle are all the same,
        # it is sufficient to look at the first quadrant and the absolute values of sin,cos:
        sin_a, cos_a = abs(math.sin(angle)), abs(math.cos(angle))
        if side_short <= 2.0 * sin_a * cos_a * side_long or abs(sin_a - cos_a) < 1e-10:
            # half constrained case: two crop corners touch the longer side,
            # the other two corners are on the mid-line parallel to the longer line
            x = 0.5 * side_short
            wr, hr = (x / sin_a, x / cos_a) if width_is_longer else (x / cos_a, x / sin_a)
        else:
            # fully constrained case: crop touches all 4 sides
            cos_2a = cos_a * cos_a - sin_a * sin_a
            wr, hr = (w * cos_a - h * sin_a) / cos_2a, (h * cos_a - w * sin_a) / cos_2a

        return dict(
            x_min=max(0, int(w / 2 - wr / 2)),
            x_max=min(w, int(w / 2 + wr / 2)),
            y_min=max(0, int(h / 2 - hr / 2)),
            y_max=min(h, int(h / 2 + hr / 2)),
        )

    check_value(angle, -360.0, 360.0)

    img_out = AFGeometric.rotate(image, angle, interpolation, border_mode, value)
    if crop_border:
        h, w = image.shape[:2]
        crop_bbox_dict = _rotated_rect_with_max_area(h, w, angle)
        x_min = crop_bbox_dict["x_min"]
        y_min = crop_bbox_dict["y_min"]
        x_max = crop_bbox_dict["x_max"]
        y_max = crop_bbox_dict["y_max"]
        img_out = AFCrops.crop(img_out, x_min, y_min, x_max, y_max)
    return img_out

random_rotate90 

random_rotate90(image: Union[ImageType, MaskType], factor: int = 0) -> Union[ImageType, MaskType]
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
110
111
112
113
114
115
def random_rotate90(
    image: Union[ImageType, MaskType],
    factor: int = 0,
) -> Union[ImageType, MaskType]:
    check_value(factor, 0, 3)
    return AF.rot90(img=image, factor=factor)

color_jitter 

color_jitter(image: ImageType, brightness: float = 1.0, contrast: float = 1.0, saturation: float = 1.0, hue: float = 0, order: List[int] = [0, 1, 2, 3]) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
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
@support_rgba
def color_jitter(
    image: ImageType,
    brightness: float = 1.0,
    contrast: float = 1.0,
    saturation: float = 1.0,
    hue: float = 0,
    order: List[int] = [0, 1, 2, 3],
) -> ImageType:
    if not AF.is_rgb_image(image) and not AF.is_grayscale_image(image):
        raise TypeError("ColorJitter transformation expects 1-channel or 3-channel images.")

    check_value(brightness, 0.01, 10.00)
    check_value(contrast, 0.01, 10.00)
    check_value(saturation, 0.01, 10.00)
    check_value(hue, -0.50, 0.50)

    transforms = [
        AF.adjust_brightness_torchvision,
        AF.adjust_contrast_torchvision,
        AF.adjust_saturation_torchvision,
        AF.adjust_hue_torchvision,
    ]
    params = [brightness, contrast, saturation, hue]

    for i in order:
        image = transforms[i](image, params[i])
    return image

blur 

blur(image: ImageType, ksize: int = 3) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
148
149
150
151
def blur(image: ImageType, ksize: int = 3) -> ImageType:
    check_value(ksize, 1, 100)

    return AF.blur(image, ksize)

gaussian_blur 

gaussian_blur(image: ImageType, ksize: int = 3, sigma: float = 0.0) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
154
155
156
157
158
159
160
161
def gaussian_blur(image: ImageType, ksize: int = 3, sigma: float = 0.0) -> ImageType:
    check_value(ksize, 1, 100)
    check_value(sigma, 0.00, 100.00)

    if ksize % 2 != 1:
        ksize = ksize + 1

    return AF.gaussian_blur(image, ksize, sigma)

adjust_brightness 

adjust_brightness(image: ImageType, brightness: float = 0.0, brightness_by_max: bool = True) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
184
185
186
187
188
189
190
191
192
def adjust_brightness(
    image: ImageType, brightness: float = 0.0, brightness_by_max: bool = True
) -> ImageType:
    check_value(brightness, -1.00, 1.00)

    alpha = 1.0
    beta = 0.0 + brightness

    return AF.brightness_contrast_adjust(image, alpha, beta, brightness_by_max)

adjust_contrast 

adjust_contrast(image: ImageType, contrast: float = 0.0) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
195
196
197
198
199
200
201
def adjust_contrast(image: ImageType, contrast: float = 0.0) -> ImageType:
    check_value(contrast, -1.00, 1.00)

    alpha = 1.0 + contrast
    beta = 0.0

    return AF.brightness_contrast_adjust(image, alpha, beta)

adjust_hue 

adjust_hue(image: ImageType, hue: float = 0.0) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
204
205
206
207
208
209
210
211
212
213
@support_gray
@support_rgba
def adjust_hue(image: ImageType, hue: float = 0.0) -> ImageType:
    check_value(hue, -1.00, 1.00)

    hue_shift = int(hue * 180)
    sat_shift = 0
    val_shift = 0

    return AF.shift_hsv(image, hue_shift, sat_shift, val_shift)

adjust_saturation 

adjust_saturation(image: ImageType, saturation: float = 0.0) -> ImageType

adjust_saturation

Parameters:

  • image (ImageType) –

    입력 이미지

  • saturation (float, default: 0.0 ) –

    변형 강도, [-1.0, 1.0] 범위, Defaults to 0.0.

Returns:

  • ImageType ( ImageType ) –

    결과 이미지

Note

Ablumentation의 AF.shift_hsv()와 다른 알고리즘을 사용합니다. AF.shift_hsv()의 경우 색이 없는 픽셀을 붉은 색으로 변형합니다. 이 함수의 경우 색이 없는 픽셀은 변형하지 않습니다.

Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
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
@support_gray
@support_rgba
def adjust_saturation(image: ImageType, saturation: float = 0.0) -> ImageType:
    """adjust_saturation

    Args:
        image (ImageType): 입력 이미지
        saturation (float, optional): 변형 강도, [-1.0, 1.0] 범위, Defaults to 0.0.

    Returns:
        ImageType: 결과 이미지

    Note:
        Ablumentation의 AF.shift_hsv()와 다른 알고리즘을 사용합니다.
        AF.shift_hsv()의 경우 색이 없는 픽셀을 붉은 색으로 변형합니다.
        이 함수의 경우 색이 없는 픽셀은 변형하지 않습니다.

    """
    check_value(saturation, -1.00, 1.00)
    saturation = (saturation + 1) ** 2
    gray = cv2.cvtColor(image, cv2.COLOR_RGB2GRAY)
    result = image * saturation + gray[..., None] * (1 - saturation)

    dtype = image.dtype
    if dtype == np.uint8:
        result = np.clip(result, 0, 255)
    elif dtype == np.uint16:
        result = np.clip(result, 0, 65535)
    result = result.astype(dtype)

    return result

adjust_gamma 

adjust_gamma(image: ImageType, gamma: float = 0.0) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
164
165
166
167
168
169
def adjust_gamma(image: ImageType, gamma: float = 0.0) -> ImageType:
    check_value(gamma, -1.0, 1.0)

    gamma = gamma + 1

    return AF.gamma_transform(image, gamma=gamma)

adjust_brightness_contrast 

adjust_brightness_contrast(image: ImageType, brightness: float = 0.0, contrast: float = 0.0, brightness_by_max: bool = True) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
172
173
174
175
176
177
178
179
180
181
def adjust_brightness_contrast(
    image: ImageType, brightness: float = 0.0, contrast: float = 0.0, brightness_by_max: bool = True
) -> ImageType:
    check_value(brightness, -1.00, 1.00)
    check_value(contrast, -1.00, 1.00)

    alpha = 1.0 + contrast
    beta = 0.0 + brightness

    return AF.brightness_contrast_adjust(image, alpha, beta, brightness_by_max)

iso_noise 

iso_noise(image: ImageType, color_shift: float = 0.05, intensity: float = 0.5, random_state: Optional[int] = None) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
249
250
251
252
253
254
255
256
257
258
259
260
@support_gray
@support_rgba
def iso_noise(
    image: ImageType,
    color_shift: float = 0.05,
    intensity: float = 0.50,
    random_state: Optional[int] = None,
) -> ImageType:
    check_value(color_shift, 0.00, 1.00)
    check_value(intensity, 0.00, 2.00)

    return AF.iso_noise(image, color_shift, intensity, np.random.RandomState(random_state))

image_compression 

image_compression(image: ImageType, quality: int = 100, image_type: str = '.jpeg') -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
263
264
265
266
267
268
269
270
@support_gray
@support_rgba
def image_compression(
    image: ImageType,
    quality: int = 100,
    image_type: str = ".jpeg",
) -> ImageType:
    return AF.image_compression(image, quality, image_type)

sharpen 

sharpen(image: ImageType, sharpening_matrix: ndarray) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
273
274
275
276
277
278
279
@support_gray
@support_rgba
def sharpen(
    image: ImageType,
    sharpening_matrix: np.ndarray,
) -> ImageType:
    return AF.convolve(image, sharpening_matrix)

multiplicative_noise 

multiplicative_noise(image: ImageType, multiplier: ndarray) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
282
283
284
285
286
287
288
@support_gray
@support_rgba
def multiplicative_noise(
    image: ImageType,
    multiplier: np.ndarray,
) -> ImageType:
    return AF.multiply(image, multiplier)

ratio_jitter 

ratio_jitter(image: ImageType, proportion_left: float = 0.0, proportion_right: float = 0.0, proportion_top: float = 0.0, proportion_bottom: float = 0.0, resampling: str = 'bilinear', border_mode: int = cv2.BORDER_CONSTANT, value: Union[int, float, List[int], List[float]] = 0) -> Union[ImageType, MaskType]
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
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
def ratio_jitter(
    image: ImageType,
    proportion_left: float = 0.0,
    proportion_right: float = 0.0,
    proportion_top: float = 0.0,
    proportion_bottom: float = 0.0,
    resampling: str = "bilinear",
    border_mode: int = cv2.BORDER_CONSTANT,
    value: Union[int, float, List[int], List[float]] = 0,
) -> Union[ImageType, MaskType]:
    check_value(proportion_left, -0.50, 0.50)
    check_value(proportion_right, -0.50, 0.50)
    check_value(proportion_top, -0.50, 0.50)
    check_value(proportion_bottom, -0.50, 0.50)

    h_original, w_original = image.shape[:2]

    left = int(w_original * proportion_left)
    right = int(w_original * proportion_right)
    top = int(h_original * proportion_top)
    bottom = int(h_original * proportion_bottom)

    # crop
    crop_left = left if left > 0 else 0
    crop_right = right if right > 0 else 0
    crop_top = top if top > 0 else 0
    crop_bottom = bottom if bottom > 0 else 0

    crop_w = max(w_original - crop_left - crop_right, 1)
    crop_h = max(h_original - crop_top - crop_bottom, 1)

    image = image[crop_top : crop_top + crop_h, crop_left : crop_left + crop_w]

    # padding
    pad_left = 0 if left > 0 else -left
    pad_right = 0 if right > 0 else -right
    pad_top = 0 if top > 0 else -top
    pad_bottom = 0 if bottom > 0 else -bottom

    image = cv2.copyMakeBorder(
        image,
        top=pad_top,
        bottom=pad_bottom,
        left=pad_left,
        right=pad_right,
        borderType=border_mode,
        value=value,
    )

    # resize
    target_size = (w_original, h_original)
    image = resize_image(image, target_size, resampling)

    return image

zoom 

zoom(image: ImageType, ratio: float = 1.0, h_start: float = 0.0, w_start: float = 0.0, resampling: str = 'bilinear', border_mode: int = cv2.BORDER_CONSTANT, value: Union[int, float, List[int], List[float]] = 0) -> Union[ImageType, MaskType]
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
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
def zoom(
    image: ImageType,
    ratio: float = 1.0,
    h_start: float = 0.0,
    w_start: float = 0.0,
    resampling: str = "bilinear",
    border_mode: int = cv2.BORDER_CONSTANT,
    value: Union[int, float, List[int], List[float]] = 0,
) -> Union[ImageType, MaskType]:
    check_value(ratio, 0.01, 100.00)
    check_value(h_start, 0.00, 1.00)
    check_value(w_start, 0.00, 1.00)

    if ratio < 1.0:
        output_image = _zoom_out(
            image=image,
            ratio=ratio,
            h_start=h_start,
            w_start=w_start,
            resampling=resampling,
            border_mode=border_mode,
            value=value,
        )
    elif ratio == 1.0:
        output_image = image
    elif ratio > 1.0:
        output_image = _zoom_in(
            image=image,
            ratio=ratio,
            h_start=h_start,
            w_start=w_start,
            resampling=resampling,
        )
    else:
        raise NotImplementedError

    return output_image

random_resized_crop 

random_resized_crop(image: ImageType, h_scale: float = 1.0, w_scale: float = 1.0, h_start: float = 0.0, w_start: float = 0.0, resampling: str = 'bilinear') -> Union[ImageType, MaskType]

Crop the given image to given scale and then resize it to original size.

Parameters:

  • image (ImageType) –

    original image

  • h_scale (float, default: 1.0 ) –

    crop height scale. Defaults to 1.0.

  • w_scale (float, default: 1.0 ) –

    crop width scale. Defaults to 1.0.

  • h_start (float, default: 0.0 ) –

    crop height start ratio. Defaults to 0.0.

  • w_start (float, default: 0.0 ) –

    crop width start ratio. Defaults to 0.0.

  • resampling (str, default: 'bilinear' ) –

    interpolation method. Defaults to "bilinear".

Returns:

Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
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
def random_resized_crop(
    image: ImageType,
    h_scale: float = 1.0,
    w_scale: float = 1.0,
    h_start: float = 0.0,
    w_start: float = 0.0,
    resampling: str = "bilinear",
) -> Union[ImageType, MaskType]:
    """Crop the given image to given scale and then resize it to original size.

    Args:
        image (ImageType): original image
        h_scale (float, optional): crop height scale. Defaults to 1.0.
        w_scale (float, optional): crop width scale. Defaults to 1.0.
        h_start (float, optional): crop height start ratio. Defaults to 0.0.
        w_start (float, optional): crop width start ratio. Defaults to 0.0.
        resampling (str, optional): interpolation method. Defaults to "bilinear".

    Returns:
        Union[ImageType, MaskType]: augmented data
    """
    check_value(h_scale, 0.01, 1.00)
    check_value(w_scale, 0.01, 1.00)
    check_value(h_start, 0.00, 1.00)
    check_value(w_start, 0.00, 1.00)

    h_original, w_original = image.shape[:2]
    target_height = int(round(h_original * h_scale))
    target_width = int(round(w_original * w_scale))

    # crop
    image = AFCrops.random_crop(
        img=image,
        crop_height=target_height,
        crop_width=target_width,
        h_start=min(h_start, 1.0 - 1e-5),
        w_start=min(w_start, 1.0 - 1e-5),
    )

    # resize
    target_size = (w_original, h_original)
    image = resize_image(image, target_size, resampling)

    return image

random_resized_crop_and_pad 

random_resized_crop_and_pad(image: ImageType, scale: float = 1.0, aspect_ratio: float = 1.0, h_start: float = 0.0, w_start: float = 0.0, height: Optional[int] = None, width: Optional[int] = None, resampling: str = 'bilinear', border_mode: int = cv2.BORDER_CONSTANT, value: Union[int, float, List[int], List[float]] = 0) -> Union[ImageType, MaskType]
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
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
def random_resized_crop_and_pad(
    image: ImageType,
    scale: float = 1.0,
    aspect_ratio: float = 1.0,
    h_start: float = 0.0,
    w_start: float = 0.0,
    height: Optional[int] = None,
    width: Optional[int] = None,
    resampling: str = "bilinear",
    border_mode: int = cv2.BORDER_CONSTANT,
    value: Union[int, float, List[int], List[float]] = 0,
) -> Union[ImageType, MaskType]:
    check_value(scale, 0.01, 1.00)
    check_value(aspect_ratio, 0.10, 10.00)
    check_value(h_start, 0.00, 1.00)
    check_value(w_start, 0.00, 1.00)

    h_original, w_original = image.shape[:2]
    area = h_original * w_original
    target_area = scale * area

    crop_height = int(round(math.sqrt(target_area / aspect_ratio)))
    crop_width = int(round(math.sqrt(target_area * aspect_ratio)))

    # pad
    pad_top = max(crop_height - h_original, 0)
    pad_bottom = max(crop_height - h_original, 0)
    pad_left = max(crop_width - w_original, 0)
    pad_right = max(crop_width - w_original, 0)

    image = cv2.copyMakeBorder(
        image,
        top=pad_top,
        bottom=pad_bottom,
        left=pad_left,
        right=pad_right,
        borderType=border_mode,
        value=value,
    )

    # crop
    image = AFCrops.random_crop(
        img=image,
        crop_height=crop_height,
        crop_width=crop_width,
        h_start=min(h_start, 1.0 - 1e-5),
        w_start=min(w_start, 1.0 - 1e-5),
    )

    # resize
    target_size = (width or w_original, height or h_original)
    image = resize_image(image, target_size, resampling)

    return image

light_reflect 

light_reflect(image: ImageType, xc: float, yc: float, x_radius: float, y_radius: float, angle: float) -> ImageType
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
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
@support_gray
def light_reflect(
    image: ImageType, xc: float, yc: float, x_radius: float, y_radius: float, angle: float
) -> ImageType:
    check_value(xc, 0.00, 1.00)
    check_value(yc, 0.00, 1.00)
    check_value(x_radius, 0.00, 1.00)
    check_value(y_radius, 0.00, 1.00)
    check_value(angle, 0.00, 360.00)

    if x_radius == 0 or y_radius == 0:
        return image

    image = cv2.cvtColor(image, cv2.COLOR_RGB2HSV)

    h_original, w_original = image.shape[:2]

    xc = ratio_to_value(xc, 0, w_original)
    yc = ratio_to_value(yc, 0, h_original)
    x_radius = ratio_to_value(x_radius, 0, w_original)
    y_radius = ratio_to_value(y_radius, 0, h_original)
    angle = math.radians(angle)

    c, r = np.meshgrid(np.arange(w_original), np.arange(h_original))
    x_rot = (c - xc) * math.cos(angle) + (r - yc) * math.sin(angle)
    y_rot = -(c - xc) * math.sin(angle) + (r - yc) * math.cos(angle)
    d = x_rot * x_rot / x_radius / x_radius + y_rot * y_rot / y_radius / y_radius
    k = 2 - np.power(d, 0.5)
    k[d >= 1] = 1

    image[:, :, 2] = np.clip(image[:, :, 2] * k, 0, 255).astype(np.uint8)

    image = cv2.cvtColor(image, cv2.COLOR_HSV2RGB)

    return image

perspective_transform 

perspective_transform(image: ImageType, offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType, interpolate_method: str = 'nearest')
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
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
618
619
620
621
622
623
624
625
def perspective_transform(
    image: ImageType,
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
    interpolate_method: str = "nearest",
):
    height, width = image.shape[:2]

    for point_offset in [
        offset_top_left,
        offset_top_right,
        offset_bottom_right,
        offset_bottom_left,
    ]:
        offset_x, offset_y = point_offset

        check_value(offset_x, 0, 49)
        check_value(offset_y, 0, 49)

    transform_matrix = calculate_transform_matrix(
        width=width,
        height=height,
        offset_top_left=offset_top_left,
        offset_top_right=offset_top_right,
        offset_bottom_right=offset_bottom_right,
        offset_bottom_left=offset_bottom_left,
    )
    image = cv2.warpPerspective(
        image, transform_matrix, (width, height), flags=INTERPOLATE_METHOD_CV2[interpolate_method]
    )

    return image

erase 

erase(image: ImageType, x: int, y: int, w: int, h: int, v: ndarray)
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
700
701
702
703
704
705
706
707
708
709
710
711
@support_gray
@support_rgba
def erase(
    image: ImageType,
    x: int,
    y: int,
    w: int,
    h: int,
    v: np.ndarray,
):
    image[y : y + h, x : x + w, ...] = v
    return image.astype("uint8")

grayscale 

grayscale(image: ImageType) -> ndarray
Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
714
715
716
717
718
719
720
@support_gray
@support_rgba
def grayscale(
    image: ImageType,
) -> np.ndarray:
    image = AF.to_gray(image)
    return image.astype("uint8")

check_range 

check_range(data: Union[Sequence[int], Sequence[float]], min_value: Union[int, float], max_value: Union[int, float])
Source code in SaigeToolkit/data/transform/function_util.py 
21
22
23
24
25
26
27
28
29
30
31
32
33
def check_range(
    data: Union[Sequence[int], Sequence[float]],
    min_value: Union[int, float],
    max_value: Union[int, float],
):
    if not (isinstance(data, Sequence) and len(data) == 2):
        raise AugmentationParameterTypeError

    check_value(data[0], min_value, max_value)
    check_value(data[1], min_value, max_value)

    if not (min_value <= data[0] <= data[1] <= max_value):
        raise AugmentationParameterRangeError

box_function

PolygonType module-attribute 

PolygonType = List[ndarray]

BBoxesType module-attribute 

BBoxesType = Union[ndarray, NumpyBoxes]

OffsetType module-attribute 

OffsetType = Tuple[float, float]

NumpyBoxes

Bases: ndarray

coordinate instance-attribute 
coordinate: str
__new__ 
__new__(input_array, coordinate: str, dtype=None) -> NumpyBoxes
Source code in SaigeToolkit/data/dataclass/box.py 
59
60
61
62
63
64
65
66
67
def __new__(cls, input_array, coordinate: str, dtype=None) -> NumpyBoxes:
    obj = np.array(input_array).view(cls)

    cls._check_boxes(obj)

    if dtype is not None:
        obj = obj.astype(dtype)
    obj.coordinate = coordinate
    return obj
__array_finalize__ 
__array_finalize__(obj)
Source code in SaigeToolkit/data/dataclass/box.py 
69
70
71
72
def __array_finalize__(self, obj):
    if obj is None:
        return
    self.coordinate = getattr(obj, "coordinate", None)
__array_function__ 
__array_function__(func, types, args, kwargs)
Source code in SaigeToolkit/data/dataclass/box.py 
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
def __array_function__(self, func, types, args, kwargs):
    def unwrap(e):
        return np.asarray(e) if isinstance(e, NumpyBoxes) else e

    def wrap(e, coordinate):
        return NumpyBoxes(e, coordinate) if isinstance(e, np.ndarray) else e

    def get_coordinate(args, kwargs):
        flat_args, _ = tree_flatten(args)
        flat_kwargs, _ = tree_flatten(kwargs)
        coordinates = [e.coordinate for e in flat_args + flat_kwargs if isinstance(e, NumpyBoxes)]

        assert len(set(coordinates)) == 1
        coordinate = coordinates[0]

        return coordinate

    kwargs = kwargs or {}
    ret = func(*tree_map(unwrap, args), **tree_map(unwrap, kwargs))
    coordinate = get_coordinate(args, kwargs)
    wrap_with_coordinate = functools.partial(wrap, coordinate=coordinate)
    ret = tree_map(wrap_with_coordinate, ret)

    return ret
convert_coordinate 
convert_coordinate(coordinate: str) -> NumpyBoxes
Source code in SaigeToolkit/data/dataclass/box.py 
 99
100
101
102
def convert_coordinate(self, coordinate: str) -> NumpyBoxes:
    new_boxes: NumpyBoxes = convert_coordinate(self, self.coordinate, coordinate)
    new_boxes.coordinate = coordinate
    return new_boxes
to_numpy 
to_numpy() -> ndarray
Source code in SaigeToolkit/data/dataclass/box.py 
104
105
def to_numpy(self) -> np.ndarray:
    return np.array(self)
to_tensor 
to_tensor() -> Tensor
Source code in SaigeToolkit/data/dataclass/box.py 
107
108
def to_tensor(self) -> torch.Tensor:
    return torch.from_numpy(self.to_numpy())
_check_boxes staticmethod 
_check_boxes(boxes: ndarray)
Source code in SaigeToolkit/data/dataclass/box.py 
110
111
112
113
114
115
116
117
118
@staticmethod
def _check_boxes(boxes: np.ndarray):
    # 숫자가 아닌 값이 들어오는 경우
    if not np.issubdtype(boxes.dtype, np.number):
        raise BoxValueError

    # box position이 음수가 들어오는 경우
    if np.any(boxes < 0):
        raise BoxValueError

rotate_polygons 

rotate_polygons(polygons: PolygonType, angle: float, image_size: Tuple[int], target_size: Optional[Tuple[int]] = None) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
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
def rotate(
    polygons: PolygonType,
    angle: float,
    image_size: Tuple[int],
    target_size: Optional[Tuple[int]] = None,
) -> PolygonType:
    if len(polygons) == 0:
        return polygons

    dtype = polygons[0].dtype
    w, h = image_size

    anchor_origin = np.array([w // 2, h // 2]).astype(dtype)

    if target_size is None:
        anchor_target = anchor_origin
    else:
        anchor_target = np.array([target_size[0] // 2, target_size[1] // 2]).astype(dtype)

    theta = np.deg2rad(angle)
    rotation_matrix = [[np.cos(theta), -np.sin(theta)], [np.sin(theta), np.cos(theta)]]

    rotated_polygons = []
    for polygon in polygons:
        # Translate polygon so that the rotation point is at the origin
        rotated_polygon = polygon - anchor_origin

        # Rotate the polygon
        rotated_polygon = np.dot(rotated_polygon, rotation_matrix)

        # Translate the polygon back
        rotated_polygon += anchor_target

        rotated_polygons.append(rotated_polygon)

    return rotated_polygons

calculate_transform_matrix 

calculate_transform_matrix(width: int, height: int, offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType) -> ndarray

Calculates the perspective transformation matrix using the given width, height, and corner points of a rectangle.

Parameters:

  • width (int) –

    Width of the original image.

  • height (int) –

    Height of the original image.

  • offset_top_left (OffsetType) –

    The offset ratio of the top-left corner point. Represented by the coordinates (x, y) and has a range of [0, 49]. Calculate xand y according to the procedure below.

    x` = width * offset_top_left[0]
y` = height * offset_top_left[1]
  • offset_bottom_right (OffsetType) –

    The offset ratio of the bottom-right corner point. Calculate xand y according to the procedure below.

    x` = width - width * offset_bottom_right[0]
y` = hegiht - height * offset_bottom_right[1]
  • offset_top_right (OffsetType) –

    The offset ratio of the top-left corner point.

  • offset_bottom_left (OffsetType) –

    The offset ratio of the bottom-left corner point.

Returns:

  • ndarray –

    np.ndarray: The 4x3 perspective transform matrix.

Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
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
692
693
694
695
696
697
def calculate_transform_matrix(
    width: int,
    height: int,
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
) -> np.ndarray:
    """
    Calculates the perspective transformation matrix
    using the given width, height, and corner points of a rectangle.

    Parameters:
        width (int): Width of the original image.
        height (int): Height of the original image.
        offset_top_left (OffsetType):
            The offset ratio of the top-left corner point.
            Represented by the coordinates (x, y) and has a range of [0, 49].
            Calculate x` and y` according to the procedure below.

            ```
            x` = width * offset_top_left[0]
            y` = height * offset_top_left[1]
            ```

        offset_bottom_right (OffsetType):
            The offset ratio of the bottom-right corner point.
            Calculate x` and y` according to the procedure below.

            ```
            x` = width - width * offset_bottom_right[0]
            y` = hegiht - height * offset_bottom_right[1]
            ```

        offset_top_right (OffsetType): The offset ratio of the top-left corner point.
        offset_bottom_left (OffsetType): The offset ratio of the bottom-left corner point.

    Returns:
        np.ndarray: The 4x3 perspective transform matrix.

    """
    points_from = np.float32([[0, 0], [width, 0], [width, height], [0, height]])

    point_top_left: OffsetType = (
        (offset_top_left[0] / 100) * width,
        (offset_top_left[1] / 100) * height,
    )
    point_top_right: OffsetType = (
        width - (offset_top_right[0] / 100) * width,
        (offset_top_right[1] / 100) * height,
    )
    point_bottom_right: OffsetType = (
        width - (offset_bottom_right[0] / 100) * width,
        height - (offset_bottom_right[1] / 100) * height,
    )
    point_bottom_left: OffsetType = (
        (offset_bottom_left[0] / 100) * width,
        height - (offset_bottom_left[1] / 100) * height,
    )
    points_to = np.float32(
        [
            point_top_left,
            point_top_right,
            point_bottom_right,
            point_bottom_left,
        ]
    )
    transform_matrix = cv2.getPerspectiveTransform(points_from, points_to)

    return transform_matrix

check_value 

check_value(data: Union[int, float], min_value: Union[int, float], max_value: Union[int, float])
Source code in SaigeToolkit/data/transform/function_util.py 
13
14
15
16
17
18
def check_value(data: Union[int, float], min_value: Union[int, float], max_value: Union[int, float]):
    if not (isinstance(data, (int, float))):
        raise AugmentationParameterTypeError

    if not (min_value <= data <= max_value):
        raise AugmentationParameterRangeError

preserve_coordinates 

preserve_coordinates(func)

Box augmentation이 (left, top, right, bottom) coordinate system을 기반으로 구현 되어있기 때문에, input bboxes의 coordinate system을 확인하고 augmentation에 맞는 coordinate system으로 변환하고, augmentation이 끝나면 다시 기존 coordinate system으로 변경하여 출력합니다.

  • np.ndarray의 경우 coordinate system을 체크할 수 없기 때문에 (left, top, right, bottom) coordinate system이라고 가정합니다.

  • NumpyBBoxes의 경우 NumpyBBoxes 내부 변수 coordinate과 내부 함수 convert_coordinate를 활용하여 구현됩니다.

Source code in SaigeToolkit/data/transform/box_function.py 
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
def preserve_coordinates(func):
    """
    Box augmentation이 (left, top, right, bottom) coordinate system을 기반으로 구현 되어있기 때문에,
    input bboxes의 coordinate system을 확인하고 augmentation에 맞는 coordinate system으로 변환하고,
    augmentation이 끝나면 다시 기존 coordinate system으로 변경하여 출력합니다.

    - np.ndarray의 경우
    coordinate system을 체크할 수 없기 때문에 (left, top, right, bottom) coordinate system이라고 가정합니다.

    - NumpyBBoxes의 경우
    NumpyBBoxes 내부 변수 coordinate과 내부 함수 convert_coordinate를 활용하여 구현됩니다.

    """
    _FUNCTIONAL_BBOX_COORDINATE = "xyxy"

    def wrapped_function_for_ndarray(bboxes: np.ndarray, *args, **kwargs) -> np.ndarray:
        new_bboxes = func(bboxes, *args, **kwargs)

        return new_bboxes

    def wrapped_function_for_numpyboxes(bboxes: NumpyBoxes, *args, **kwargs) -> NumpyBoxes:
        # original coordinate -> (left, top, right, bottom)
        original_coordinate = bboxes.coordinate  # record original coordinate
        bboxes = bboxes.convert_coordinate(_FUNCTIONAL_BBOX_COORDINATE)  # -> (left, top, right, bottom)

        # box augmentation
        new_bboxes = func(bboxes, *args, **kwargs)
        if not isinstance(new_bboxes, bboxes.__class__):
            new_bboxes = bboxes.__class__(new_bboxes, _FUNCTIONAL_BBOX_COORDINATE)

        # (left, top, right, bottom) -> original coordinate
        new_bboxes = new_bboxes.convert_coordinate(original_coordinate)

        return new_bboxes

    @wraps(func)
    def wrapped_function(bboxes: BBoxesType, *args, **kwargs) -> BBoxesType:
        # isinstance가 상속된 type도 true를 return 하기 때문에 여기서는 type(instance) == class로 체크
        if isinstance(bboxes, NumpyBoxes):
            new_bboxes = wrapped_function_for_numpyboxes(bboxes, *args, **kwargs)
        elif isinstance(bboxes, np.ndarray):
            new_bboxes = wrapped_function_for_ndarray(bboxes, *args, **kwargs)
        else:
            raise NotImplementedError

        return new_bboxes

    return wrapped_function

resize_box 

resize_box(bboxes: BBoxesType, image_size: Tuple[int], tw: int, th: int) -> BBoxesType

bbox resize from (w, h) to (tw, th)

Source code in SaigeToolkit/data/transform/box_function.py 
62
63
64
65
66
67
68
69
70
71
72
@preserve_coordinates
def resize_box(bboxes: BBoxesType, image_size: Tuple[int], tw: int, th: int) -> BBoxesType:
    """bbox resize from (w, h) to (tw, th)"""
    dtype = bboxes.dtype
    w, h = image_size
    if w == tw and h == th:
        return bboxes
    else:
        w_scale = tw / w
        h_scale = th / h
        return (bboxes * (w_scale, h_scale, w_scale, h_scale)).astype(dtype)

crop_box 

crop_box(bboxes: BBoxesType, cropping_box: Union[ndarray, List[int]]) -> BBoxesType

bbox crop. An image is cropped at (new_left, new_top, new_right, new_bottom)

Source code in SaigeToolkit/data/transform/box_function.py 
75
76
77
78
79
80
81
82
83
84
@preserve_coordinates
def crop_box(bboxes: BBoxesType, cropping_box: Union[np.ndarray, List[int]]) -> BBoxesType:
    """bbox crop. An image is cropped at (new_left, new_top, new_right, new_bottom)"""
    dtype = bboxes.dtype
    new_left, new_top, new_right, new_bottom = map(round, cropping_box)
    new_h = new_bottom - new_top
    new_bboxes = bboxes - (new_left, new_top, new_left, new_top)
    new_bboxes[:, [0, 1]] = np.maximum(new_bboxes[:, [0, 1]], 0)
    new_bboxes[:, [2, 3]] = np.minimum(new_bboxes[:, [2, 3]], (new_right - new_left, new_h))
    return new_bboxes.astype(dtype)

hflip_box 

hflip_box(bboxes: BBoxesType, image_size: Tuple[int]) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
87
88
89
90
91
92
@preserve_coordinates
def hflip_box(bboxes: BBoxesType, image_size: Tuple[int]) -> BBoxesType:
    w, h = image_size
    bboxes = bboxes.copy()
    bboxes[:, [0, 2]] = w - bboxes[:, [2, 0]]
    return bboxes

vflip_box 

vflip_box(bboxes: BBoxesType, image_size: Tuple[int]) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
 95
 96
 97
 98
 99
100
@preserve_coordinates
def vflip_box(bboxes: BBoxesType, image_size: Tuple[int]) -> BBoxesType:
    w, h = image_size
    bboxes = bboxes.copy()
    bboxes[:, [1, 3]] = h - bboxes[:, [3, 1]]
    return bboxes

rotate 

rotate(bboxes: BBoxesType, angle: Union[float, int], image_size: Tuple[int], target_size: Optional[Tuple[int]] = None, clipping: bool = True) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
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
@preserve_coordinates
def rotate(
    bboxes: BBoxesType,
    angle: Union[float, int],
    image_size: Tuple[int],
    target_size: Optional[Tuple[int]] = None,
    clipping: bool = True,
) -> BBoxesType:
    dtype = bboxes.dtype
    w, h = image_size

    def _bboxes_to_polygons(bboxes: BBoxesType) -> PolygonType:
        polygons = []
        for bbox in bboxes:
            x1, y1, x2, y2 = bbox
            polygon = np.array([[x1, y1], [x1, y2], [x2, y2], [x2, y1]]).astype(dtype)
            polygons.append(polygon)
        return polygons

    def _polygons_to_bboxes(polygons: PolygonType) -> BBoxesType:
        bboxes = []
        for polygon in polygons:
            # Extract the bounding box from the rotated polygon
            x1, y1 = np.min(polygon, axis=0)
            x2, y2 = np.max(polygon, axis=0)
            bboxes.append([x1, y1, x2, y2])

        bboxes = np.array(bboxes).reshape(-1, 4)
        return bboxes

    if bboxes.size == 0:
        return bboxes

    polygons = _bboxes_to_polygons(bboxes)
    rotated_polygons = rotate_polygons(
        polygons=polygons, angle=angle, image_size=image_size, target_size=target_size
    )
    rotated_bboxes = _polygons_to_bboxes(rotated_polygons)

    if clipping:
        rotated_bboxes[:, [0, 2]] = np.clip(rotated_bboxes[:, [0, 2]], 0, w - 1)
        rotated_bboxes[:, [1, 3]] = np.clip(rotated_bboxes[:, [1, 3]], 0, h - 1)
    return rotated_bboxes.astype(dtype)

rotate90 

rotate90(bboxes: BBoxesType, factor: int, image_size: Tuple[int], clipping: bool = True) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
@preserve_coordinates
def rotate90(
    bboxes: BBoxesType,
    factor: int,
    image_size: Tuple[int],
    clipping: bool = True,
) -> BBoxesType:
    angle = factor * 90.0
    if factor % 2 == 0:
        target_size = image_size
    else:
        target_size = (image_size[1], image_size[0])
    return rotate(
        bboxes=bboxes, angle=angle, image_size=image_size, target_size=target_size, clipping=clipping
    )

translate_box 

translate_box(bboxes: BBoxesType, offset: Tuple[int], image_size: Tuple[int]) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
165
166
167
168
169
170
171
172
173
174
@preserve_coordinates
def translate_box(bboxes: BBoxesType, offset: Tuple[int], image_size: Tuple[int]) -> BBoxesType:
    w, h = image_size
    x_offset, y_offset = offset
    new_bboxes = bboxes.copy()
    new_bboxes[:, 0] = np.maximum(np.minimum(bboxes[:, 0] - x_offset, w), 0)
    new_bboxes[:, 1] = np.maximum(np.minimum(bboxes[:, 1] - y_offset, h), 0)
    new_bboxes[:, 2] = np.minimum(np.maximum(bboxes[:, 2] - x_offset, 0), w)
    new_bboxes[:, 3] = np.minimum(np.maximum(bboxes[:, 3] - y_offset, 0), h)
    return new_bboxes

ratio_jitter 

ratio_jitter(bboxes: BBoxesType, image_size: Tuple[int], proportion_left: float = 0.0, proportion_right: float = 0.0, proportion_top: float = 0.0, proportion_bottom: float = 0.0) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
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
@preserve_coordinates
def ratio_jitter(
    bboxes: BBoxesType,
    image_size: Tuple[int],
    proportion_left: float = 0.0,
    proportion_right: float = 0.0,
    proportion_top: float = 0.0,
    proportion_bottom: float = 0.0,
) -> BBoxesType:
    w, h = image_size

    left = int(w * proportion_left)
    right = int(w * proportion_right)
    top = int(h * proportion_top)
    bottom = int(h * proportion_bottom)

    # Crop
    crop_left = left if left > 0 else 0
    crop_right = right if right > 0 else 0
    crop_top = top if top > 0 else 0
    crop_bottom = bottom if bottom > 0 else 0

    crop_w = max(w - crop_left - crop_right, 1)
    crop_h = max(h - crop_top - crop_bottom, 1)

    bboxes = crop_box(bboxes, [crop_left, crop_top, crop_left + crop_w, crop_top + crop_h])

    # Padding
    pad_left = 0 if left > 0 else -left
    pad_right = 0 if right > 0 else -right
    pad_top = 0 if top > 0 else -top
    pad_bottom = 0 if bottom > 0 else -bottom

    bboxes = resize_box(bboxes, (crop_w + pad_left + pad_right, crop_h + pad_top + pad_bottom), w, h)

    # Translate
    w_scale = (crop_w + pad_left + pad_right) / w
    h_scale = (crop_h + pad_top + pad_bottom) / h
    bboxes = translate_box(bboxes, [-pad_left / w_scale, -pad_top / h_scale], image_size)

    return bboxes

zoom 

zoom(bboxes: BBoxesType, image_size: Tuple[int], ratio: float = 1.0, h_start: float = 0.0, w_start: float = 0.0) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
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
@preserve_coordinates
def zoom(
    bboxes: BBoxesType,
    image_size: Tuple[int],
    ratio: float = 1.0,
    h_start: float = 0.0,
    w_start: float = 0.0,
) -> BBoxesType:
    w, h = image_size
    if ratio == 1.0:
        return bboxes
    target_size = (int(w * ratio), int(h * ratio))
    bboxes = resize_box(bboxes, image_size, target_size[0], target_size[1])
    if ratio > 1.0:
        h_start = min(h_start, 1.0 - 1e-5)
        w_start = min(w_start, 1.0 - 1e-5)
        x1 = int((target_size[0] - w + 1) * w_start)
        y1 = int((target_size[1] - h + 1) * h_start)
        bboxes = crop_box(bboxes, [x1, y1, x1 + w, y1 + h])
    elif ratio < 1.0:
        x1 = int((w - target_size[0] + 1) * w_start)
        y1 = int((h - target_size[1] + 1) * h_start)
        bboxes = translate_box(bboxes, [-x1, -y1], image_size)

    return bboxes

random_resized_crop_and_pad 

random_resized_crop_and_pad(bboxes: BBoxesType, image_size: Tuple[int], scale: float = 1.0, aspect_ratio: float = 1.0, h_start: float = 0.0, w_start: float = 0.0, height: Optional[int] = None, width: Optional[int] = None) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
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
@preserve_coordinates
def random_resized_crop_and_pad(
    bboxes: BBoxesType,
    image_size: Tuple[int],
    scale: float = 1.0,
    aspect_ratio: float = 1.0,
    h_start: float = 0.0,
    w_start: float = 0.0,
    height: Optional[int] = None,
    width: Optional[int] = None,
) -> BBoxesType:
    w_original, h_original = image_size
    area = h_original * w_original
    target_area = scale * area

    crop_height = int(round(math.sqrt(target_area / aspect_ratio)))
    crop_width = int(round(math.sqrt(target_area * aspect_ratio)))

    w_start = min(w_start, 1.0 - 1e-5)
    h_start = min(h_start, 1.0 - 1e-5)

    # Padding
    pad_top = max(crop_height - h_original, 0)
    pad_bottom = max(crop_height - h_original, 0)
    pad_left = max(crop_width - w_original, 0)
    pad_right = max(crop_width - w_original, 0)

    padded_image_size = [w_original + pad_left + pad_right, h_original + pad_top + pad_bottom]
    bboxes = translate_box(bboxes, [-pad_left, -pad_top], padded_image_size)

    # Crop
    x1 = int((padded_image_size[0] - crop_width + 1) * w_start)
    y1 = int((padded_image_size[1] - crop_height + 1) * h_start)
    bboxes = crop_box(bboxes, [x1, y1, x1 + crop_width, y1 + crop_height])

    # Resize
    bboxes = resize_box(
        bboxes,
        [crop_width, crop_height],
        width or w_original,
        height or h_original,
    )

    return bboxes

perspective_transform 

perspective_transform(bboxes: BBoxesType, image_size: Tuple[int], offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType) -> BBoxesType
Source code in SaigeToolkit/data/transform/box_function.py 
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
@preserve_coordinates
def perspective_transform(
    bboxes: BBoxesType,
    image_size: Tuple[int],
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
) -> BBoxesType:
    width, height = image_size

    for point_offset in [
        offset_top_left,
        offset_top_right,
        offset_bottom_right,
        offset_bottom_left,
    ]:
        offset_x, offset_y = point_offset

        check_value(offset_x, 0, 49)
        check_value(offset_y, 0, 49)

    transform_matrix = calculate_transform_matrix(
        width=width,
        height=height,
        offset_top_left=offset_top_left,
        offset_top_right=offset_top_right,
        offset_bottom_right=offset_bottom_right,
        offset_bottom_left=offset_bottom_left,
    )

    new_bboxes = bboxes.copy()
    for idx_box, bbox in enumerate(bboxes):
        # [(x1, y1), (x2, y1), (x2, y2), (x1, y2)]
        points = [(bbox[0], bbox[1]), (bbox[2], bbox[1]), (bbox[2], bbox[3]), (bbox[0], bbox[3])]

        dtype_max_value = np.iinfo(np.int32).max
        min_x, min_y = dtype_max_value, dtype_max_value
        max_x, max_y = 0, 0

        for point in points:
            x, y, scaling_factor = np.dot(transform_matrix, np.append(point, 1))
            x = int(x / scaling_factor)
            y = int(y / scaling_factor)

            if x < min_x:
                min_x = x

            elif x > max_x:
                max_x = x

            if y < min_y:
                min_y = y

            elif y > max_y:
                max_y = y

        new_bboxes[idx_box] = [min_x, min_y, max_x, max_y]

    return new_bboxes

polygon_function

PolygonType module-attribute 

PolygonType = List[ndarray]

OffsetType module-attribute 

OffsetType = Tuple[float, float]

calculate_transform_matrix 

calculate_transform_matrix(width: int, height: int, offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType) -> ndarray

Calculates the perspective transformation matrix using the given width, height, and corner points of a rectangle.

Parameters:

  • width (int) –

    Width of the original image.

  • height (int) –

    Height of the original image.

  • offset_top_left (OffsetType) –

    The offset ratio of the top-left corner point. Represented by the coordinates (x, y) and has a range of [0, 49]. Calculate xand y according to the procedure below.

    x` = width * offset_top_left[0]
y` = height * offset_top_left[1]
  • offset_bottom_right (OffsetType) –

    The offset ratio of the bottom-right corner point. Calculate xand y according to the procedure below.

    x` = width - width * offset_bottom_right[0]
y` = hegiht - height * offset_bottom_right[1]
  • offset_top_right (OffsetType) –

    The offset ratio of the top-left corner point.

  • offset_bottom_left (OffsetType) –

    The offset ratio of the bottom-left corner point.

Returns:

  • ndarray –

    np.ndarray: The 4x3 perspective transform matrix.

Source code in SaigeToolkit/data/transform/augmentation/augment_function.py 
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
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
692
693
694
695
696
697
def calculate_transform_matrix(
    width: int,
    height: int,
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
) -> np.ndarray:
    """
    Calculates the perspective transformation matrix
    using the given width, height, and corner points of a rectangle.

    Parameters:
        width (int): Width of the original image.
        height (int): Height of the original image.
        offset_top_left (OffsetType):
            The offset ratio of the top-left corner point.
            Represented by the coordinates (x, y) and has a range of [0, 49].
            Calculate x` and y` according to the procedure below.

            ```
            x` = width * offset_top_left[0]
            y` = height * offset_top_left[1]
            ```

        offset_bottom_right (OffsetType):
            The offset ratio of the bottom-right corner point.
            Calculate x` and y` according to the procedure below.

            ```
            x` = width - width * offset_bottom_right[0]
            y` = hegiht - height * offset_bottom_right[1]
            ```

        offset_top_right (OffsetType): The offset ratio of the top-left corner point.
        offset_bottom_left (OffsetType): The offset ratio of the bottom-left corner point.

    Returns:
        np.ndarray: The 4x3 perspective transform matrix.

    """
    points_from = np.float32([[0, 0], [width, 0], [width, height], [0, height]])

    point_top_left: OffsetType = (
        (offset_top_left[0] / 100) * width,
        (offset_top_left[1] / 100) * height,
    )
    point_top_right: OffsetType = (
        width - (offset_top_right[0] / 100) * width,
        (offset_top_right[1] / 100) * height,
    )
    point_bottom_right: OffsetType = (
        width - (offset_bottom_right[0] / 100) * width,
        height - (offset_bottom_right[1] / 100) * height,
    )
    point_bottom_left: OffsetType = (
        (offset_bottom_left[0] / 100) * width,
        height - (offset_bottom_left[1] / 100) * height,
    )
    points_to = np.float32(
        [
            point_top_left,
            point_top_right,
            point_bottom_right,
            point_bottom_left,
        ]
    )
    transform_matrix = cv2.getPerspectiveTransform(points_from, points_to)

    return transform_matrix

check_value 

check_value(data: Union[int, float], min_value: Union[int, float], max_value: Union[int, float])
Source code in SaigeToolkit/data/transform/function_util.py 
13
14
15
16
17
18
def check_value(data: Union[int, float], min_value: Union[int, float], max_value: Union[int, float]):
    if not (isinstance(data, (int, float))):
        raise AugmentationParameterTypeError

    if not (min_value <= data <= max_value):
        raise AugmentationParameterRangeError

vertical_flip 

vertical_flip(polygons: PolygonType, image_size: Tuple[int]) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
11
12
13
14
15
16
def vertical_flip(polygons: PolygonType, image_size: Tuple[int]) -> PolygonType:
    _, h = image_size
    polygons = deepcopy(polygons)
    for polygon in polygons:
        polygon[:, 1] = h - polygon[:, 1]
    return polygons

horizontal_flip 

horizontal_flip(polygons: PolygonType, image_size: Tuple[int]) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
19
20
21
22
23
24
def horizontal_flip(polygons: PolygonType, image_size: Tuple[int]) -> PolygonType:
    w, _ = image_size
    polygons = deepcopy(polygons)
    for polygon in polygons:
        polygon[:, 0] = w - polygon[:, 0]
    return polygons

rotate 

rotate(polygons: PolygonType, angle: float, image_size: Tuple[int], target_size: Optional[Tuple[int]] = None) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
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
def rotate(
    polygons: PolygonType,
    angle: float,
    image_size: Tuple[int],
    target_size: Optional[Tuple[int]] = None,
) -> PolygonType:
    if len(polygons) == 0:
        return polygons

    dtype = polygons[0].dtype
    w, h = image_size

    anchor_origin = np.array([w // 2, h // 2]).astype(dtype)

    if target_size is None:
        anchor_target = anchor_origin
    else:
        anchor_target = np.array([target_size[0] // 2, target_size[1] // 2]).astype(dtype)

    theta = np.deg2rad(angle)
    rotation_matrix = [[np.cos(theta), -np.sin(theta)], [np.sin(theta), np.cos(theta)]]

    rotated_polygons = []
    for polygon in polygons:
        # Translate polygon so that the rotation point is at the origin
        rotated_polygon = polygon - anchor_origin

        # Rotate the polygon
        rotated_polygon = np.dot(rotated_polygon, rotation_matrix)

        # Translate the polygon back
        rotated_polygon += anchor_target

        rotated_polygons.append(rotated_polygon)

    return rotated_polygons

rotate90 

rotate90(polygons: PolygonType, factor: int, image_size: Tuple[int]) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
65
66
67
68
69
70
71
72
73
74
75
def rotate90(
    polygons: PolygonType,
    factor: int,
    image_size: Tuple[int],
) -> PolygonType:
    angle = factor * 90.0
    if factor % 2 == 0:
        target_size = image_size
    else:
        target_size = (image_size[1], image_size[0])
    return rotate(polygons=polygons, angle=angle, image_size=image_size, target_size=target_size)

ratio_jitter 

ratio_jitter(polygons: PolygonType, image_size: Tuple[int], proportion_left: float, proportion_right: float, proportion_top: float, proportion_bottom: float) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
 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
def ratio_jitter(
    polygons: PolygonType,
    image_size: Tuple[int],
    proportion_left: float,
    proportion_right: float,
    proportion_top: float,
    proportion_bottom: float,
) -> PolygonType:
    w, h = image_size

    left = int(w * proportion_left)
    right = int(w * proportion_right)
    top = int(h * proportion_top)
    bottom = int(h * proportion_bottom)

    # Crop
    crop_left = left if left > 0 else 0
    crop_right = right if right > 0 else 0
    crop_top = top if top > 0 else 0
    crop_bottom = bottom if bottom > 0 else 0

    crop_w = max(w - crop_left - crop_right, 1)
    crop_h = max(h - crop_top - crop_bottom, 1)
    polygons = translate_polygon(polygons, [crop_left, crop_top])

    # Padding
    pad_left = 0 if left > 0 else -left
    pad_right = 0 if right > 0 else -right
    pad_top = 0 if top > 0 else -top
    pad_bottom = 0 if bottom > 0 else -bottom

    polygons = resize_polygon(
        polygons, (crop_w + pad_left + pad_right, crop_h + pad_top + pad_bottom), w, h
    )

    # Translate
    w_scale = (crop_w + pad_left + pad_right) / w
    h_scale = (crop_h + pad_top + pad_bottom) / h
    polygons = translate_polygon(polygons, [-int(pad_left / w_scale), -int(pad_top / h_scale)])

    return polygons

zoom 

zoom(polygons: PolygonType, image_size: Tuple[int], ratio: float, h_start: float, w_start: float) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
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
def zoom(
    polygons: PolygonType,
    image_size: Tuple[int],
    ratio: float,
    h_start: float,
    w_start: float,
) -> PolygonType:
    w, h = image_size

    if ratio == 1.0:
        return polygons

    target_size = (int(w * ratio), int(h * ratio))
    polygons = resize_polygon(polygons, image_size, target_size[0], target_size[1])

    x1 = 0
    y1 = 0

    if ratio > 1.0:
        h_start = min(h_start, 1.0 - 1e-5)
        w_start = min(w_start, 1.0 - 1e-5)
        x1 = int((target_size[0] - w + 1) * w_start)
        y1 = int((target_size[1] - h + 1) * h_start)

    elif ratio < 1.0:
        x1 = -int((w - target_size[0] + 1) * w_start)
        y1 = -int((h - target_size[1] + 1) * h_start)

    polygons = translate_polygon(polygons, [x1, y1])

    return polygons

random_resized_crop_and_pad 

random_resized_crop_and_pad(polygons: PolygonType, image_size: Tuple[int], scale: float, aspect_ratio: float, h_start: float, w_start: float, h_target: Optional[int] = None, w_target: Optional[int] = None, **params) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
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
def random_resized_crop_and_pad(
    polygons: PolygonType,
    image_size: Tuple[int],
    scale: float,
    aspect_ratio: float,
    h_start: float,
    w_start: float,
    h_target: Optional[int] = None,
    w_target: Optional[int] = None,
    **params
) -> PolygonType:
    w_original, h_original = image_size
    area = h_original * w_original
    target_area = scale * area

    crop_height = int(round(math.sqrt(target_area / aspect_ratio)))
    crop_width = int(round(math.sqrt(target_area * aspect_ratio)))

    # Padding
    pad_top = max(crop_height - h_original, 0)
    pad_bottom = max(crop_height - h_original, 0)
    pad_left = max(crop_width - w_original, 0)
    pad_right = max(crop_width - w_original, 0)

    padded_image_size = [w_original + pad_left + pad_right, h_original + pad_top + pad_bottom]
    polygons = translate_polygon(
        polygons,
        [-(pad_left), -(pad_top)],
    )

    # Crop
    x1 = int((padded_image_size[0] - crop_width + 1) * w_start)
    y1 = int((padded_image_size[1] - crop_height + 1) * h_start)
    polygons = translate_polygon(polygons, [x1, y1])

    # Resize
    polygons = resize_polygon(
        polygons,
        [crop_width, crop_height],
        w_target or w_original,
        h_target or h_original,
    )
    return polygons

perspective_transform 

perspective_transform(polygons: PolygonType, image_size: Tuple[int], offset_top_left: OffsetType, offset_top_right: OffsetType, offset_bottom_right: OffsetType, offset_bottom_left: OffsetType) -> PolygonType
Source code in SaigeToolkit/data/transform/polygon_function.py 
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
def perspective_transform(
    polygons: PolygonType,
    image_size: Tuple[int],
    offset_top_left: OffsetType,
    offset_top_right: OffsetType,
    offset_bottom_right: OffsetType,
    offset_bottom_left: OffsetType,
) -> PolygonType:
    width, height = image_size

    for point_offset in [
        offset_top_left,
        offset_top_right,
        offset_bottom_right,
        offset_bottom_left,
    ]:
        offset_x, offset_y = point_offset

        check_value(offset_x, 0, 49)
        check_value(offset_y, 0, 49)

    transform_matrix = calculate_transform_matrix(
        width=width,
        height=height,
        offset_top_left=offset_top_left,
        offset_top_right=offset_top_right,
        offset_bottom_right=offset_bottom_right,
        offset_bottom_left=offset_bottom_left,
    )
    new_polygons = deepcopy(polygons)
    for idx_poly, polygon in enumerate(polygons):
        for idx_pt, point in enumerate(polygon):
            x, y, scaling_factor = np.dot(transform_matrix, np.append(point, 1))

            new_polygons[idx_poly][idx_pt][0] = int(x / scaling_factor)
            new_polygons[idx_poly][idx_pt][1] = int(y / scaling_factor)

    return new_polygons

resize_polygon 

resize_polygon(polygons: PolygonType, image_size: Tuple[int], tw: int, th: int) -> PolygonType

Resize polygon from (w, h) to (tw, th)

Source code in SaigeToolkit/data/transform/polygon_function.py 
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
def resize_polygon(polygons: PolygonType, image_size: Tuple[int], tw: int, th: int) -> PolygonType:
    """Resize polygon from (w, h) to (tw, th)"""
    if len(polygons) == 0:
        return polygons

    w, h = image_size
    if w == tw and h == th:
        return polygons

    dtype = polygons[0].dtype

    w_scale = tw / w
    h_scale = th / h

    new_polygons = deepcopy(polygons)

    for polygon in new_polygons:
        polygon[:, 0] = polygon[:, 0] * w_scale
        polygon[:, 1] = polygon[:, 1] * h_scale
        polygon = polygon.astype(dtype)

    return new_polygons

translate_polygon 

translate_polygon(polygons: PolygonType, offset: Tuple[int]) -> PolygonType

Translate polyfon from [(x1, y1), ... ] to [(x1 - x_offset), (y1 - y_offset), ...]

Source code in SaigeToolkit/data/transform/polygon_function.py 
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
def translate_polygon(
    polygons: PolygonType,
    offset: Tuple[int],
) -> PolygonType:
    """Translate polyfon from [(x1, y1), ... ] to [(x1 - x_offset), (y1 - y_offset), ...]"""
    if len(polygons) == 0:
        return polygons

    if offset[0] == 0 and offset[1] == 0:
        return polygons

    dtype = polygons[0].dtype

    x_offset, y_offset = np.array(offset).astype(dtype)

    new_polygons = deepcopy(polygons)

    for polygon in new_polygons:
        polygon[:, 0] -= x_offset
        polygon[:, 1] -= y_offset

    return new_polygons