How to Compute The Area of a Set of Bounding Boxes in PyTorch?

In this article, we are going to see how to compute the area of a set of bounding boxes in PyTorch. We can compute the area of a set of bounding boxes by using the box_area() method of torchvision.io module.

box_area() method

This method accepts bounding boxes as an input and returns the area of the given bounding boxes. The input bounding boxes must be torch Tensors with [N,4] size, where N represents the number of bounding boxes for which the area will be computed. The bounding boxes are expected to be in the format (x_min, y_min, x_max, y_max), where 0 ≤ x_min < x_max, and 0 ≤ y_min < y_max. Before computing the area of a bounding box we use unsqueeze to make this bounding box tensor into a 2D tensor.

Syntax: torchvision.ops.box_area(boxes)

Parameter:

• boxes: This method accepts bounding boxes as input.

Return: This method return area for each box.

Stepwise Implementation

Step 1: Import the required libraries.

Python

import torch
import torchvision
from torchvision.io import read_image
from torchvision.utils import draw_bounding_boxes
from torchvision.ops import box_area

Step 2: Read the input image from your computer.

Python

img = read_image('img.png')

Step 3: define a bounding box and convert this box into a torch tensor.

Python

b_box = [80, 70, 500, 200]
b_box = torch.tensor(b_box, dtype=torch.int)

Step 4: unsqueeze the given bounding box to make it a 2D tensor. Execute this step only if we want to compute the area of a single bounding box else skip this step.

Python

b_box = b_box.unsqueeze(0)

Step 5: Compute the above defined bounding box area and store this computed area in a variable for further use.

Python

area = box_area(b_box)

Step 6: set this computed area on the label.

Python

label = [f"b_box area = {area.item()}"]

Step 7: Draw a bounding box on the image and put the above-defined label on box.

Python

img = draw_bounding_boxes(img, b_box, labels=label,
                          width=4, colors=(255, 0, 0))

Step 8: transform this image into a PIL image

Python

img = torchvision.transforms.ToPILImage()(img)

Step 9: Display the output image.

Python

img.show()

The below image is used for demonstration:

Example 1:

in this example, we are computing the area of a single bounding box and set this computed area as a label.

Python

# Import the required libraries
import torch
import torchvision
from torchvision.io import read_image
from torchvision.utils import draw_bounding_boxes
from torchvision.ops import box_area

# read input image from your computer
img = read_image('img.png')

# bounding box are xmin, ymin, xmax, ymax
b_box = [80, 70, 500, 200]

# convert the bounding box to torch tensor
b_box = torch.tensor(b_box, dtype=torch.int)

# unsqueeze the given bounding box to make
# it 2D tensor
b_box = b_box.unsqueeze(0)

# Compute the bounding box area
area = box_area(b_box)

# set this computed area on label
label = [f"b_box area = {area.item()}"]

# draw the above define bounding box on image
# Set the above define label on image
img = draw_bounding_boxes(img, b_box, labels=label,
                          width=4, colors=(255, 0, 0))


# transform this image to PIL image
img = torchvision.transforms.ToPILImage()(img)

# display result
img.show()

Output:

Example 2:

in this example, we are computing the area of multiple bounding boxes and set this computed area as a label for each box.

Python

# Import the required libraries
import torch
from PIL import Image
import torchvision
from torchvision.io import read_image
from torchvision.utils import draw_bounding_boxes
from torchvision.ops import box_area

# read input image from your computer
img = read_image('img.png')

# create boxes
b_box1 = [80, 70, 500, 200]
b_box2 = [80, 230, 500, 300]
b_box3 = [580, 70, 720, 300]
b_box = [b_box1, b_box2,b_box3]

# convert the bounding box to torch tensor
b_box = torch.tensor(b_box, dtype=torch.int)

# Compute the bounding box area
area = box_area(b_box)

# set this computed area on label
labels = [f"b_box area ={n}" for n in area]

# draw the above define bounding boxes on image
img=draw_bounding_boxes(img, b_box, labels = labels, width=4,
                        colors=["orange", "white","red"])

# transform this image to PIL image
img = torchvision.transforms.ToPILImage()(img)

# display result
img.show()

Output: