Skip to main content

​
Overview

The draw_boxes module provides utilities for visualizing object detection results by drawing bounding boxes on images with class labels and confidence scores.

​
Functions

​
draw_boxes()

Draw bounding boxes with class names and optional scores on an image. 
def draw_boxes(image, boxes, box_classes, class_names, scores=None)

​
Parameters

​
image
numpy.ndarray
required
An array of shape (width, height, 3) with pixel values in the range [0, 1]. The image will be converted to PIL format internally.
​
boxes
numpy.ndarray
required
An array of shape (num_boxes, 4) containing bounding box coordinates in the format (y_min, x_min, y_max, x_max).
​
box_classes
list
required
A list of integer indices into the class_names array, indicating which class each box belongs to.
​
class_names
list
required
A list of string class names (e.g., ['person', 'car', 'dog']).
​
scores
numpy.ndarray
default:"None"
An optional array of confidence scores for each box. When provided, scores are displayed alongside class names in the format “Class 0.95”.

​
Returns

​
image
numpy.ndarray
A copy of the input image (as a numpy array) with bounding boxes, labels, and scores drawn on it.

​
Description

The function performs the following operations:
  1. Converts the input image from normalized float values to PIL Image format
  2. Loads the FiraMono-Medium font for text rendering
  3. Generates distinct colors for each class using HSV color space
  4. Draws rectangles for each bounding box with the corresponding class color
  5. Adds text labels with class names and optional scores
  6. Returns the modified image as a numpy array
Visual Features:
  • Box thickness scales with image size (approximately 1/300 of combined width and height)
  • Font size adapts to image height (3% of image height)
  • Labels are positioned above boxes when space permits, otherwise below
  • Each class receives a consistent, vibrant color

​
Example Usage

import numpy as np
from PIL import Image
from yad2k.utils.draw_boxes import draw_boxes

# Load an image
image = Image.open('test.jpg')
image_array = np.array(image) / 255.0  # Normalize to [0, 1]

# Detection results
boxes = np.array([
    [100, 50, 200, 150],   # y_min, x_min, y_max, x_max
    [150, 200, 250, 300]
])
box_classes = [0, 1]  # Indices into class_names
class_names = ['person', 'car']
scores = np.array([0.95, 0.87])

# Draw boxes on image
result = draw_boxes(image_array, boxes, box_classes, class_names, scores)

# Save or display result
Image.fromarray(result).save('output.jpg')

​
Example Without Scores

# Draw boxes without confidence scores
result = draw_boxes(image_array, boxes, box_classes, class_names)
# Labels will show only class names: "person", "car"

​
get_colors_for_classes()

Generate a list of random but consistent colors for object classes. 
def get_colors_for_classes(num_classes)

​
Parameters

​
num_classes
int
required
The number of distinct classes to generate colors for.

​
Returns

​
colors
list
A list of RGB tuples (R, G, B) with values in the range [0, 255]. The list contains num_classes colors.

​
Description

This function generates visually distinct colors using HSV color space and caches them for consistency across multiple calls. Features:
  • Deterministic: Uses a fixed random seed (10101) for reproducible colors
  • Cached: Returns previously generated colors if num_classes matches
  • Well-distributed: Generates colors evenly spaced in HSV hue space
  • Shuffled: Colors are shuffled to decorrelate adjacent class indices

​
Example Usage

from yad2k.utils.draw_boxes import get_colors_for_classes

# Generate colors for 80 COCO classes
colors = get_colors_for_classes(80)

print(colors[0])  # (205, 231, 0) - RGB tuple
print(len(colors))  # 80

# Subsequent calls with same num_classes return cached colors
colors2 = get_colors_for_classes(80)
print(colors == colors2)  # True

​
Notes

  • The draw_boxes() function requires the FiraMono-Medium.otf font file to be located at font/FiraMono-Medium.otf relative to the working directory
  • Input images should be normalized to [0, 1] range before passing to draw_boxes()
  • Box coordinates follow the YOLO format: (y_min, x_min, y_max, x_max)
  • The function prints each box’s label and coordinates to stdout for debugging

Build docs developers (and LLMs) love

Get started for freeTalk to us