Bounding boxes are rectangles that mark objects on an image. There are multiple formats of bounding boxes annotations. Each format uses its specific representation of bouning boxes coordinates. Albumentations supports four formats:
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
3,
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
4,
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
5, and
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
6 .
Let's take a look at each of those formats and how they represent coordinates of bounding boxes.
As an example, we will use an image from the dataset named Common Objects in Context. It contains one bounding box that marks a cat. The image width is 640 pixels, and its height is 480 pixels. The width of the bounding box is 322 pixels, and its height is 117 pixels.
The bounding box has the following
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
7 coordinates of its corners: top-left is
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
8 or
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
9, top-right is
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
0 or
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
1, bottom-left is
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
2 or
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
3, bottom-right is
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
4 or
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
5. As you see, coordinates of the bounding box's corners are calculated with respect to the top-left corner of the image which has
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
7 coordinates
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
7.
pascal_voc
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
3 is a format used by the Pascal VOC dataset. Coordinates of a bounding box are encoded with four values in pixels:
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
9.
image = cv2.imread["/path/to/image.jpg"]
image = cv2.cvtColor[image, cv2.COLOR_BGR2RGB]
0 and
image = cv2.imread["/path/to/image.jpg"]
image = cv2.cvtColor[image, cv2.COLOR_BGR2RGB]
1 are coordinates of the top-left corner of the bounding box.
image = cv2.imread["/path/to/image.jpg"]
image = cv2.cvtColor[image, cv2.COLOR_BGR2RGB]
2 and
image = cv2.imread["/path/to/image.jpg"]
image = cv2.cvtColor[image, cv2.COLOR_BGR2RGB]
3 are coordinates of bottom-right corner of the bounding box.
Coordinates of the example bounding box in this format are
image = cv2.imread["/path/to/image.jpg"]
image = cv2.cvtColor[image, cv2.COLOR_BGR2RGB]
4.
albumentations
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
4 is similar to
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
3, because it also uses four values
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
9 to represent a bounding box. But unlike
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
3,
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
4 uses normalized values. To normalize values, we divide coordinates in pixels for the x- and y-axis by the width and the height of the image.
Coordinates of the example bounding box in this format are
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
0 which are
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
1.
Albumentations uses this format internally to work with bounding boxes and augment them.
coco
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
5 is a format used by the Common Objects in Context COCO dataset.
In
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
5, a bounding box is defined by four values in pixels
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
4. They are coordinates of the top-left corner along with the width and height of the bounding box.
Coordinates of the example bounding box in this format are
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
5.
yolo
In
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
6, a bounding box is represented by four values
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
7.
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
8 and
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
9 are the normalized coordinates of the center of the bounding box. To make coordinates normalized, we take pixel values of x and y, which marks the center of the bounding box on the x- and y-axis. Then we divide the value of x by the width of the image and value of y by the height of the image.
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
0 and
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
1 represent the width and the height of the bounding box. They are normalized as well.
Coordinates of the example bounding box in this format are
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
2 which are
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
3.
Bounding boxes augmentation
Just like with images and masks augmentation, the process of augmenting bounding boxes consists of 4 steps.
- You import the required libraries.
- You define an augmentation pipeline.
- You read images and bounding boxes from the disk.
- You pass an image and bounding boxes to the augmentation pipeline and receive augmented images and boxes.
Note
Some transforms in Albumentation don't support bounding boxes. If you try to use them you will get an exception. Please refer to this article to check whether a transform can augment bounding boxes.
Step 1. Import the required libraries.
import albumentations as A
import cv2
Step 2. Define an augmentation pipeline.
Here an example of a minimal declaration of an augmentation pipeline that works with bounding boxes.
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
Note that unlike image and masks augmentation,
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
4 now has an additional parameter
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
5. You need to pass an instance of
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
6 to that argument.
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
6 specifies settings for working with bounding boxes.
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
8 sets the format for bounding boxes coordinates.
It can either be
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
3,
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
4,
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
5 or
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
6. This value is required because Albumentation needs to know the coordinates' source format for bounding boxes to apply augmentations correctly.
Besides
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
8,
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
6 supports a few more settings.
Here is an example of
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
4 that shows all available settings with
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
6:
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco', min_area=1024, min_visibility=0.1, label_fields=['class_labels']]]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
7 and
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
8
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
7 and
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
8 parameters control what Albumentations should do to the augmented bounding boxes if their size has changed after augmentation. The size of bounding boxes could change if you apply spatial augmentations, for example, when you crop a part of an image or when you resize an image.
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
7 is a value in pixels. If the area of a bounding box after augmentation becomes smaller than
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
7, Albumentations will drop that box. So the returned list of augmented bounding boxes won't contain that bounding box.
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
8 is a value between 0 and 1. If the ratio of the bounding box area after augmentation to
transformed = transform[image=image, bboxes=bboxes]
transformed_image = transformed['image']
transformed_bboxes = transformed['bboxes']
4 becomes smaller than
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
8, Albumentations will drop that box. So if the augmentation process cuts the most of the bounding box, that box won't be present in the returned list of the augmented bounding boxes.
Here is an example image that contains two bounding boxes. Bounding boxes coordinates are declared using the
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
5 format.
First, we apply the
transformed = transform[image=image, bboxes=bboxes]
transformed_image = transformed['image']
transformed_bboxes = transformed['bboxes']
7 augmentation without declaring parameters
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
7 and
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
8. The augmented image contains two bounding boxes.
Next, we apply the same
transformed = transform[image=image, bboxes=bboxes]
transformed_image = transformed['image']
transformed_bboxes = transformed['bboxes']
7 augmentation, but now we also use the
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
7 parameter. Now, the augmented image contains only one bounding box, because the other bounding box's area after augmentation became smaller than
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
7, so Albumentations dropped that bounding box.
Finally, we apply the
transformed = transform[image=image, bboxes=bboxes]
transformed_image = transformed['image']
transformed_bboxes = transformed['bboxes']
7 augmentation with the
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
8. After that augmentation, the resulting image doesn't contain any bounding box, because visibility of all bounding boxes after augmentation are below threshold set by
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
8.
Class labels for bounding boxes
Besides coordinates, each bounding box should have an associated class label that tells which object lies inside the bounding box. There are two ways to pass a label for a bounding box.
Let's say you have an example image with three objects:
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
6,
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
7, and
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
8. Bounding boxes coordinates in the
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
5 format for those objects are
class_labels = ['cat', 'dog', 'parrot']
0,
class_labels = ['cat', 'dog', 'parrot']
1, and
class_labels = ['cat', 'dog', 'parrot']
2.
1. You can pass labels along with bounding boxes coordinates by adding them as additional values to the list of coordinates.
For the image above, bounding boxes with class labels will become
class_labels = ['cat', 'dog', 'parrot']
3,
class_labels = ['cat', 'dog', 'parrot']
4, and
class_labels = ['cat', 'dog', 'parrot']
5.
Class labels could be of any type: integer, string, or any other Python data type. For example, integer values as class labels will look the following:
class_labels = ['cat', 'dog', 'parrot']
6,
class_labels = ['cat', 'dog', 'parrot']
7, and
class_labels = ['cat', 'dog', 'parrot']
8
Also, you can use multiple class values for each bounding box, for example
class_labels = ['cat', 'dog', 'parrot']
9,
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
00, and
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
01.
2.You can pass labels for bounding boxes as a separate list [the preferred way].
For example, if you have three bounding boxes like
class_labels = ['cat', 'dog', 'parrot']
0,
class_labels = ['cat', 'dog', 'parrot']
1, and
class_labels = ['cat', 'dog', 'parrot']
2 you can create a separate list with values like
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
05, or
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
06 that contains class labels for those bounding boxes. Next, you pass that list with class labels as a separate argument to the
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
07 function. Albumentations needs to know the names of all those lists with class labels to join them with augmented bounding boxes correctly. Then, if a bounding box is dropped after augmentation because it is no longer visible, Albumentations will drop the class label for that box as well. Use
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
08 parameter to set names for all arguments in
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
07 that will contain label descriptions for bounding boxes [more on that in Step 4].
Step 3. Read images and bounding boxes from the disk.
Read an image from the disk.
image = cv2.imread["/path/to/image.jpg"]
image = cv2.cvtColor[image, cv2.COLOR_BGR2RGB]
Bounding boxes can be stored on the disk in different serialization formats: JSON, XML, YAML, CSV, etc. So the code to read bounding boxes depends on the actual format of data on the disk.
After you read the data from the disk, you need to prepare bounding boxes for Albumentations.
Albumentations expects that bounding boxes will be represented as a list of lists. Each list contains information about a single bounding box. A bounding box definition should have at list four elements that represent the coordinates of that bounding box. The actual meaning of those four values depends on the format of bounding boxes [either
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
3,
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
4,
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
5, or
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
6]. Besides four coordinates, each definition of a bounding box may contain one or more extra values. You can use those extra values to store additional information about the bounding box, such as a class label of the object inside the box. During augmentation, Albumentations will not process those extra values. The library will return them as is along with the updated coordinates of the augmented bounding box.
Step 4. Pass an image and bounding boxes to the augmentation pipeline and receive augmented images and boxes.
As discussed in Step 2, there are two ways of passing class labels along with bounding boxes coordinates:
1. Pass class labels along with coordinates.
So, if you have coordinates of three bounding boxes that look like this:
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
you can add a class label for each bounding box as an additional element of the list along with four coordinates. So now a list with bounding boxes and their coordinates will look the following:
bboxes = [
]
[23, 74, 295, 388, 'dog'],
[377, 294, 252, 161, 'cat'],
[333, 421, 49, 49, 'sports ball'],
or with multiple labels per each bounding box:
bboxes = [
]
[23, 74, 295, 388, 'dog', 'animal'],
[377, 294, 252, 161, 'cat', 'animal'],
[333, 421, 49, 49, 'sports ball', 'item'],
You can use any data type for declaring class labels. It can be string, integer, or any other Python data type.
Next, you pass an image and bounding boxes for it to the
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
07 function and receive the augmented image and bounding boxes.
transformed = transform[image=image, bboxes=bboxes]
transformed_image = transformed['image']
transformed_bboxes = transformed['bboxes']
2. Pass class labels in a separate argument to
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
07 [the preferred way].
Let's say you have coordinates of three bounding boxes
bboxes = [
]
[23, 74, 295, 388],
[377, 294, 252, 161],
[333, 421, 49, 49],
You can create a separate list that contains class labels for those bounding boxes:
class_labels = ['cat', 'dog', 'parrot']
Then you pass both bounding boxes and class labels to
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
07. Note that to pass class labels, you need to use the name of the argument that you declared in
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
08 when creating an instance of Compose in step 2. In our case, we set the name of the argument to
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
18.
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
0
Note that
transform = A.Compose[[
], bbox_params=A.BboxParams[format='coco']]
A.RandomCrop[width=450, height=450],
A.HorizontalFlip[p=0.5],
A.RandomBrightnessContrast[p=0.2],
08 expects a list, so you can set multiple fields that contain labels for your bounding boxes. So if you declare Compose like
What is Xywh format?
The xywh[] CSS function creates a rectangle using the specified distances from the left [ x ] and top [ y ] edges of the containing block and the specified width [ w ] and height [ h ] of the rectangle. It is a basic shape function of the data type.
How to convert bounding box x1 y1 x2 y2 to yolo style?
To convert between your [x, y] coordinates and yolo [u, v] coordinates you need to transform your data as u = x / XMAX and y = y / YMAX where XMAX , YMAX are the maximum coordinates for the image array you are using. This all depends on the image arrays being oriented the same way.
How do you normalize a bounding box coordinate?
To normalize values, we divide coordinates in pixels for the x- and y-axis by the width and the height of the image. Coordinates of the example bounding box in this format are [98 / 640, 345 / 480, 420 / 640, 462 / 480] which are [0.153125, 0.71875, 0.65625, 0.9625] .
What is the format for bounding box coordinates?
These bounding box coordinates are usually in the format of [xmin, ymin, xmax, ymax]. With these coordinates, you can easily calculate the width and height of the detected object.