Parameterized Transforms

About the Package

• The package provides a uniform, modular, and easily extendable implementation of torchvision-based transforms that provides access to their parameterization.

• With this access, the transforms enable users to achieve the following two important functionalities–

  • Given an image, return an augmentation along with the parameters used for the augmentation.

  • Given an image and augmentation parameters, return the corresponding augmentation.

Installation

• To install the package directly, run the following commands:

git clone git@github.com:apple/parameterized-transforms.git
cd parameterized-transforms
pip install -e .

• To install the package via pip, run the following command:

pip install --upgrade git+https://git@github.com:apple/parameterized-transforms.git

Get Started

• To understand the structure of parameterized transforms and the details of the package, we recommend the reader to start with the Tutorial Series.

• Otherwise, for a quick starter, check out Parameterized Transforms in a Nutshell.

Important Links

Quick-Start

• Parameterized Transforms in a Nutshell

Tutorial Series

• About the Package
• The Structure of Parametrized Transforms
• How to Write Your Own Transforms
• A Brief Introduction to the Transforms in this Package
• Parametrized Transforms in Action
• Migrate To-and-From torch/torchvision in Three Easy Steps

Python API Reference

• core
• transforms
• utils
• wrappers