Lightweight package for managing bounding boxes that works seamlessly with most computing frameworks.

CI/CD
Meta
Package

---

This package provides a simple API for managing bounding boxes. It allows you to perform transformation to your Numpy's ndarray, JAX's Array, PyTorch Tensor or Tensorflow Tensor from one orientation to another.

️️⚙️ Installation

Install the package from the PyPI registry. You may need to specify the framework that you want to managed. Numpy is enabled by default.

pip install anyboxes
# or
pip install "anyboxes[torch]" # or jax, tensorflow

Install the package from the latest commit of the repository.

pip install git+https://github.com/VDuchauffour/anyboxes

⚡ Usage

Example

from anyboxes import TorchBoxes
import torch

detections = torch.randint(0, 1000, (10, 4))

boxes = TorchBoxes.from_bottom_left_corner(detections)
boxes = boxes.to_center()
boxes.as_tensor

In a nutshell, using a Boxes involve 3 stages:

#	Stage	Methods that can be used
1	Instantiate a Boxes object with one of the from classmethods	from_top_left_corner, from_bottom_left_corner, from_two_corners, from_center
2	Apply a transformation with a to inplace methods	to_top_left_corner, to_bottom_left_corner, to_two_corners, to_center
3	Retrieve the modified data with one of the as properties	as_dict, as_tuple, as_numpy, as_array, as_tf_tensor, as_tensor

To be more specific, when a Boxes is instantiated, the following attribute are created:

Attribute	Purpose
corners_coordinates	A tuple of coordinates from top to bottom and from left to right
center_coordinates	A object that contains center coordinates
size	A object that contains width and height attributes
origin	Origin of the coordinates, can be equal to top-left or bottom-left

⛏️ Development

Clone the project

git clone https://github.com/VDuchauffour/anyboxes

In order to install all development dependencies, run the following command:

pip install -e ".[all,dev]"

To ensure that you follow the development workflow, please setup the pre-commit hooks:

pre-commit install

🧭 Roadmap

• Implementations
  • PyTorch
  • Numpy
  • JAX
  • Tensorflow
• Dispatch implementation using a metaclass