Spaces:
Runtime error
Runtime error
| import random | |
| import cv2 | |
| import numpy as np | |
| from albumentations import DualTransform, ImageOnlyTransform | |
| from albumentations.augmentations.crops.functional import crop | |
| #from albumentations.augmentations.functional import crop | |
| def isotropically_resize_image(img, size, interpolation_down=cv2.INTER_AREA, interpolation_up=cv2.INTER_CUBIC): | |
| h, w = img.shape[:2] | |
| if max(w, h) == size: | |
| return img | |
| if w > h: | |
| scale = size / w | |
| h = h * scale | |
| w = size | |
| else: | |
| scale = size / h | |
| w = w * scale | |
| h = size | |
| interpolation = interpolation_up if scale > 1 else interpolation_down | |
| resized = cv2.resize(img, (int(w), int(h)), interpolation=interpolation) | |
| return resized | |
| class IsotropicResize(DualTransform): | |
| def __init__(self, max_side, interpolation_down=cv2.INTER_AREA, interpolation_up=cv2.INTER_CUBIC, | |
| always_apply=False, p=1): | |
| super(IsotropicResize, self).__init__(always_apply, p) | |
| self.max_side = max_side | |
| self.interpolation_down = interpolation_down | |
| self.interpolation_up = interpolation_up | |
| def apply(self, img, interpolation_down=cv2.INTER_AREA, interpolation_up=cv2.INTER_CUBIC, **params): | |
| return isotropically_resize_image(img, size=self.max_side, interpolation_down=interpolation_down, | |
| interpolation_up=interpolation_up) | |
| def apply_to_mask(self, img, **params): | |
| return self.apply(img, interpolation_down=cv2.INTER_NEAREST, interpolation_up=cv2.INTER_NEAREST, **params) | |
| def get_transform_init_args_names(self): | |
| return ("max_side", "interpolation_down", "interpolation_up") | |
| class Resize4xAndBack(ImageOnlyTransform): | |
| def __init__(self, always_apply=False, p=0.5): | |
| super(Resize4xAndBack, self).__init__(always_apply, p) | |
| def apply(self, img, **params): | |
| h, w = img.shape[:2] | |
| scale = random.choice([2, 4]) | |
| img = cv2.resize(img, (w // scale, h // scale), interpolation=cv2.INTER_AREA) | |
| img = cv2.resize(img, (w, h), | |
| interpolation=random.choice([cv2.INTER_CUBIC, cv2.INTER_LINEAR, cv2.INTER_NEAREST])) | |
| return img | |
| class RandomSizedCropNonEmptyMaskIfExists(DualTransform): | |
| def __init__(self, min_max_height, w2h_ratio=[0.7, 1.3], always_apply=False, p=0.5): | |
| super(RandomSizedCropNonEmptyMaskIfExists, self).__init__(always_apply, p) | |
| self.min_max_height = min_max_height | |
| self.w2h_ratio = w2h_ratio | |
| def apply(self, img, x_min=0, x_max=0, y_min=0, y_max=0, **params): | |
| cropped = crop(img, x_min, y_min, x_max, y_max) | |
| return cropped | |
| def targets_as_params(self): | |
| return ["mask"] | |
| def get_params_dependent_on_targets(self, params): | |
| mask = params["mask"] | |
| mask_height, mask_width = mask.shape[:2] | |
| crop_height = int(mask_height * random.uniform(self.min_max_height[0], self.min_max_height[1])) | |
| w2h_ratio = random.uniform(*self.w2h_ratio) | |
| crop_width = min(int(crop_height * w2h_ratio), mask_width - 1) | |
| if mask.sum() == 0: | |
| x_min = random.randint(0, mask_width - crop_width + 1) | |
| y_min = random.randint(0, mask_height - crop_height + 1) | |
| else: | |
| mask = mask.sum(axis=-1) if mask.ndim == 3 else mask | |
| non_zero_yx = np.argwhere(mask) | |
| y, x = random.choice(non_zero_yx) | |
| x_min = x - random.randint(0, crop_width - 1) | |
| y_min = y - random.randint(0, crop_height - 1) | |
| x_min = np.clip(x_min, 0, mask_width - crop_width) | |
| y_min = np.clip(y_min, 0, mask_height - crop_height) | |
| x_max = x_min + crop_height | |
| y_max = y_min + crop_width | |
| y_max = min(mask_height, y_max) | |
| x_max = min(mask_width, x_max) | |
| return {"x_min": x_min, "x_max": x_max, "y_min": y_min, "y_max": y_max} | |
| def get_transform_init_args_names(self): | |
| return "min_max_height", "height", "width", "w2h_ratio" | |