class HorizontalBlurrer(ImageModifierABC):
max_kernel_size: int = 11
def modify(self, image: Image, intensity: float) -> Image:
kernel = self._make_zero_kernel(intensity)
self._fill_middle_row_with_ones(kernel)
self._normalize_kernel(kernel)
return self._apply_kernel(image, kernel)
def _make_zero_kernel(self, intensity: float) -> np.ndarray:
kernel_size = int(self.max_kernel_size * intensity) + 1
return np.zeros((kernel_size, kernel_size))
@staticmethod
def _fill_middle_row_with_ones(kernel: np.ndarray):
kernel[len(kernel) // 2, :] = 1
@staticmethod
def _normalize_kernel(kernel: np.ndarray):
kernel /= len(kernel)
@staticmethod
def _apply_kernel(image: Image, kernel: np.ndarray) -> Image:
return cv2.filter2D(image, -1, kernel)
if __name__ == "__main__":
from research.dataset.perturbations.utils import simple_modifier_demo
simple_modifier_demo(HorizontalBlurrer())
from dataclasses import dataclass
import cv2
from polystar.models.image import Image
from research.dataset.perturbations.image_modifiers.image_modifier_abc import ImageModifierABC
@dataclass
class GaussianBlurrer(ImageModifierABC):
max_factor: float = 0.015
def modify(self, image: Image, intensity: float) -> Image:
blur_factor = intensity * self.max_factor
width, height, *_ = image.shape
x, y = _to_odd_number(width * blur_factor), _to_odd_number(height * blur_factor)
image = cv2.GaussianBlur(image, (x, y), cv2.BORDER_DEFAULT)
return image
def _to_odd_number(number):
return int(number // 2 * 2) - 1
if __name__ == "__main__":
from research.dataset.perturbations.utils import simple_modifier_demo
simple_modifier_demo(GaussianBlurrer())
from dataclasses import dataclass
import numpy as np
from polystar.models.image import Image
from research.dataset.perturbations.image_modifiers.image_modifier_abc import ImageModifierABC
@dataclass
class BrightnessModifier(ImageModifierABC):
max_offset: float = 10.0
def modify(self, image: Image, intensity: float) -> Image:
offset = self.max_offset * intensity
perturbed_image = np.clip((image.astype(np.uint16) + offset), 0, 255).astype(np.uint8)
return perturbed_image
if __name__ == "__main__":
from research.dataset.perturbations.utils import simple_modifier_demo
simple_modifier_demo(BrightnessModifier())
from dataclasses import dataclass
import numpy as np
from polystar.models.image import Image
from research.dataset.perturbations.image_modifiers.image_modifier_abc import ImageModifierABC
@dataclass
class ContrastModifier(ImageModifierABC):
min_coef: float = 0.8
max_coef: float = 1.5
def modify(self, image: Image, intensity: float) -> Image:
coef = self.min_coef + (self.max_coef - self.min_coef) * intensity
perturbed_image = np.clip((image.astype(np.uint16) * coef), 0,
255).astype(np.uint8)
return perturbed_image
if __name__ == "__main__":
from research.dataset.perturbations.utils import simple_modifier_demo
simple_modifier_demo(ContrastModifier())
from dataclasses import dataclass
import cv2
import numpy as np
from polystar.models.image import Image
from research.dataset.perturbations.image_modifiers.image_modifier_abc import ImageModifierABC
@dataclass
class SaturationModifier(ImageModifierABC):
max_saturation: float = 0.6
def modify(self, image: Image, intensity: float) -> Image:
saturation_factor = 1 + self.max_saturation * intensity
image_hsv = cv2.cvtColor(image, cv2.COLOR_RGB2HSV)
(h, s, v) = cv2.split(image_hsv)
new_s = np.clip((s.astype(np.uint16) * saturation_factor), 0, 255).astype(np.uint8)
new_image_hsv = cv2.merge([h, new_s, v])
image_rgb = cv2.cvtColor(new_image_hsv, cv2.COLOR_HSV2RGB).astype(np.uint8)
return image_rgb
if __name__ == "__main__":
from research.dataset.perturbations.utils import simple_modifier_demo
simple_modifier_demo(SaturationModifier())
from dataclasses import dataclass
import numpy as np
from polystar.models.image import Image
from research.dataset.perturbations.image_modifiers.image_modifier_abc import ImageModifierABC
@dataclass
class GaussianNoiser(ImageModifierABC):
max_variance: float = 300.0
def modify(self, image: Image, intensity: float) -> Image:
variance = self.max_variance * intensity
sigma = variance**0.5
row, column, ch = image.shape
gaussian = np.random.normal(0, sigma, (row, column, ch))
perturbed_image = np.clip((image.astype(np.uint16) + gaussian), 0,
255).astype(np.uint8)
return perturbed_image
if __name__ == "__main__":
from research.dataset.perturbations.utils import simple_modifier_demo
simple_modifier_demo(GaussianNoiser())