Image Bounding Box

Latest Release
License

IMBO is a library to plot pretty bounding boxes with a simple Python API.
Please notice this lib does not do object detection for you, but only helps to display pretty bounding boxes with a carefully chosen set of colors.

Inspiration

The core idea came from nalepae/bounding-box All the credits goes to Manu NALEPA for the core functionalities, IMBO is built on top of bounding-box with more enhanced features keeping customization at focus.

Installation

>> pip install imbo

API

Add the bounding box and label on an image.

import cv2
import imbo
image = cv2.imread("example.jpg")
image = imbo.draw(image, left, top, right, bottom, label, color)

This method takes 5 mandatory parameters:

• image: A numpy array, channel last (ie. height x width x colors) with channels in BGR order (same as openCV format).
• left: A integer representing the left side of the bounding box.
• top: A integer representing the top side of the bounding box.
• right: A integer representing the right side of the bounding box.
• bottom: A integer representing the bottom side of the bounding box.

It also takes 8 optional parameters:

• label: A string representing the label of the bounding box, If not specified, then no label will be displayed.

• font_name: You can use multiple fonts but the .ttf file must be present in the module path.

  • Available font are: Aclonica, Courgette, Gloria, Kalam, Kaushan, Merienda, Roboto-Medium, Typewriter, Ubuntu-B

• font_size: It can be used to adjust the label text size in case if required, default=20.

• thickness: The thickness of bounding boxes, default=2.

• adjust_label: Incase while increasing the font_size or thickness the label text might get misplaced, use adjust_label=(x, y) to adjust the position.

• rescale: There are time when the input image is either of very less resolution or very high resolution in such cases use rescale=True , see this example to know how rescaling affects the output.

• rescale_width: Width to use when applying rescaling, default=1920

• rescale_height: Width to use when applying rescaling, default=1920

• bbox_color: RGB tuple, HEX code or a string representing the color of the bounding box.

• label_color: RGB tuple, HEX code or a string representing the color of the label text.

  • Possible string values are:navy, blue, aqua, teal, olive, green, lime, yellow, orange, red, maroon, fuchsia, purple, black, gray ,silver , white.

---

Rescale image without messing with original bounding box coordinates.

import cv2
import imbo

image = cv2.imread("car.jpg")
car = (88,50,827,591)
license_plate = (337, 499, 598, 561)

coordinates = [car, license_plate]
image, coordinates = imbo.rescale(image, coordinates, width=900, height=600, padding='replicate')

This method takes 3 mandatory parameters:

• image: A numpy array, channel last (ie. height x width x colors) with channels in BGR order (same as openCV format).
• coord_list: A list containing bbox tuples in order of (left, top, right, bottom).
• width/height(any one): This parameter is used to resize image and bbox accordingly while maintaining aspect ratio. Note:- if both parameters are passed at once the image will be force resized to given dimensions.

And it takes 3 optional parameters:

• keep_ratio: A Boolean to tell whether to maintain aspect ratio or not while resizing images and rescaling corresponding bounding boxes. useful when both height and width parameter are passed at once.

• padding: String representing padding type to apply, possible values are :

  • replicate, reflect, reflect, wrap, constant

• padding_color: BGR tuple, when using padding='constant' it can be used to set padding color.

  Here is an example showing how to do custom rescaling of images and bounding boxes.

Video Demo

Image Examples

The script to plot examples of this README is available here. go in root of this git repository then write:

Below are some examples of bounding boxes plotted with this library.

python docs/examples.py

credits

Again, many thanks to @nalepae for coming up with the original idea and providing some core functionalities for IMBO. Icon made by Freepik