simplebbox 0.0.10

Basic assumptions

Simple data structures are used: lists, tuples, numpy arrays. They fit well to the output of the deep learning models. Speed and simplicity is chosen over type and range safety. For better type safety one can use alternative implementations with classes. See alternatives page of the docs.

Conversion operations between integer and float numbers are sometimes cumbersome. We are trying to keep it aligned with standard python. For example rounding of float numbers while computing a center of a bounding box is done using Round_half_to_even algorithm.

Several functions for different handing of floats/ints are provided for the same operation. E.g. cxcywh_to_x0y0wh_float, cxcywh_to_x0y0wh_int

In simple use cases like a single conversion from one format to another one can probably chose any of the versions for float conversions. More thorough selection is needed in case of multiple back and forth conversions.

Installation

pip install simplebbox

Usage example

Examples of conversions:

from simplebbox.array import x0y0wh_to_x0y0x1y1, x0y0x1y1_to_x0y0wh

# convert bbox represented as list in ltwh format:
x0y0wh_to_x0y0x1y1([100, 200, 10, 20])          # [100, 200, 110, 220]

# and convert it back:
x0y0x1y1_to_x0y0wh([100, 200, 110, 220])        # [100, 200, 10, 20]

Notes

• Free software: MIT license

• Documentation: https://simplebbox.readthedocs.io.

0.0.9 (2021-03-02)

• implementation of batch processing for torch and numpy

• simplification of float/int conversions, merged some tests for array/tensor implementations

• reworked a conversion with integer divisions

Interfaces (and names) of array processing functions have changed. This is a breaking change.

0.0.7 (2020-12-17)

• Added conversion between relative and absolute coordinates.

0.0.6 (2020-12-17)

• Implemented basic conversions. Updated documentation, tests.

0.0.1 (2020-12-16)

• First release on PyPI.