Albumentation

Albumentation

所有实现的变换。

变换以及支持的类型。此处

Classification

1. 创建图片扩展流程

使用 Compose 类定义数据增强的流程。Compose 类接受一个包含所有图片变换的列表。

transform = A.Compose([
    A.RandomCrop(width=256, height=256),
    A.HorizontalFlip(p=0.5),
    A.RandomBrightnessContrast(p=0.2),
])

2. 读取图片

• Albumentation 采用 RGB 格式

3. 传入图片，获取变换后图片

# 返回是字典
transformed_image = transform(image=image)["image"]

Segmentation

需要同时变换图片和 Mask 。1，2 两布同 Classification 相同，第三步

3. 传入图片和 Mask

transformed = transform(image=image， mask=mask)
transformed_image = transformed['image']
transformed_mask = transformed['mask']

如果传入多个 Mask，采用 transform(image=image， masks=masks，masks 是 mask 组成的列表。

Detection

四种 bbox 格式：

1. pascal_voc：[x_min, y_min, x_max, y_max]
2. albumentations：Normalized [x_min, y_min, x_max, y_max]
3. coco：[x_min, y_min, width, height]
4. yolo ：Normalized [x_center, y_center, width, height]

具体流程：

1. Augmentaion Pipeline

transform = A.Compose([
    A.RandomCrop(width=450, height=450),
    A.HorizontalFlip(p=0.5),
    A.RandomBrightnessContrast(p=0.2),
], bbox_params=A.BboxParams(format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']))

min_area is a value in pixels. If the area of a bounding box after augmentation becomes smaller than min_area, Albumentations will drop that box.

min_visibility is a value between 0 and 1. If the ratio of the bounding box area after augmentation to the area of the bounding box before augmentation becomes smaller than min_visibility, Albumentations will drop that box.

label_fields 确定哪些传入参数作为 bbox 的类别。

2. Pass Image and labels

下面这个例子中，每个 bbox 有两个类（class_labels，class_categories）。

transform = A.Compose([
    A.RandomCrop(width=450, height=450),
    A.HorizontalFlip(p=0.5),
    A.RandomBrightnessContrast(p=0.2),
], bbox_params=A.BboxParams(format='coco', label_fields=['class_labels', 'class_categories'])))

class_labels = ['cat', 'dog', 'parrot']
class_categories = ['animal', 'animal', 'item']

transformed = transform(image=image, bboxes=bboxes, class_labels=class_labels, class_categories=class_categories) # 传入参数由 `Compose` 中 `label_fields` 决定
transformed_image = transformed['image']
transformed_bboxes = transformed['bboxes']
transformed_class_labels = transformed['class_labels']
transformed_class_categories = transformed['class_categories']

KeyPoints Augmentation

支持的关键点格式，此处

1. Augmentation Pipeline

transform = A.Compose([
    A.RandomCrop(width=330, height=330),
    A.RandomBrightnessContrast(p=0.2),
], keypoint_params=A.KeypointParams(format='xy', label_fields=['class_labels'], remove_invisible=True, angle_in_degrees=True))

label_fields：In some computer vision tasks, keypoints have not only coordinates but associated labels as well. (different kinds of keypoints)