powerboxes 0.2.0

Utility functions to manipulate and compute metrics on boxes

PowerBoxes

Powerboxes is a package containing utility functions for transforming bounding boxes and computing metrics. It is implemented in Rust with python bindings. It shows a significant speedup over the equivalent numpy implementations in Python, or other libraries such as shapely.

Installation

pip install powerboxes

Python Usage

import powerboxes as pb
import numpy as np

# Create a bounding box
box = np.array([[0, 0, 1, 1]])

# Compute the area of the box
area = pb.box_areas(box)

# Compute the intersection of the box with itself
intersection = pb.iou_distance(box, box)

Functions available

Box Transformations and utilities

• box_areas: Compute the area of list of boxes
• box_convert: Convert a box from one format to another. Supported formats are xyxy, xywh, cxcywh.
• remove_small_boxes: Remove boxes with area smaller than a threshold
• mask_to_boxes: Convert a mask to a list of boxes

Box Metrics

• iou_distance: Compute the intersection over union matrix of two sets of boxes
• parallel_iou_distance: Compute the intersection over union matrix of two sets of boxes in parallel
• giou_distance: Compute the generalized intersection over union matrix of two sets of boxes
• parallel_giou_distance: Compute the generalized intersection over union matrix of two sets of boxes in parallel

Box NMS

• nms: Non-maximum suppression, returns the indices of the boxes to keep