boxlib 0.3.1

Utilities for bounding box manipulation.

boxlib

A lightweight Python library for bounding box manipulation, designed for computer vision tasks.

Installation

pip install boxlib

Box Convention

All bounding boxes are NumPy arrays with format [x, y, width, height], where (x, y) is the top-left corner.

import numpy as np
import boxlib

box = np.array([10, 20, 100, 50])  # x=10, y=20, width=100, height=50

Quick Examples

import numpy as np
import boxlib

box = np.array([10, 20, 100, 50])

# Get center point
boxlib.center(box)  # array([60., 45.])

# Expand box by 2x around its center
boxlib.expand(box, 2)  # array([-40., -5., 200., 100.])

# Make box square (using max dimension)
boxlib.expand_to_square(box)  # array([10., -5., 100., 100.])

# Intersection of two boxes
box2 = np.array([50, 40, 100, 50])
boxlib.intersection(box, box2)  # array([50., 40., 60., 30.])

# IoU (Intersection over Union)
boxlib.iou(box, box2)  # 0.189...

# Crop image to bounding box
image = np.zeros((200, 200, 3))
cropped = boxlib.crop_image(image, box)  # shape: (50, 100, 3)

Functions

Geometry

• center(box) - Get center point
• corners(box) - Get four corner coordinates
• side_midpoints(box) - Get midpoints of each side
• inscribed_ellipse_points(box, n_angles, n_radii) - Sample points inside inscribed ellipse
• area(box) - Calculate area

Transformations

• shift(box, delta) - Translate box by offset
• expand(box, factor) - Scale box around its center
• expand_to_square(box) - Expand to square using max dimension
• crop_to_square(box) - Crop to square using min dimension
• box_around(center, size) - Create box from center point and size

Set Operations

• intersection(box1, box2) - Intersection of two boxes
• box_hull(box1, box2) - Smallest box containing both
• contains(box, points) - Check if points are inside box

Metrics

• iou(box1, box2) - Intersection over Union
• giou(box1, box2) - Generalized IoU

Construction

• bb_of_points(points) - Bounding box of point set
• bb_of_mask(mask) - Bounding box of binary mask
• full(imshape=None, imsize=None) - Full image bounding box
• empty() - Zero-size box at origin

Image Operations

• crop_image(image, box, pad=True) - Crop image to box with optional padding

License

MIT