Lightnet

During my PhD, I developed a plethora of detection models, each with their pros & cons. All of these models, together with their necessary loss, pre- and post-processing have been implemented in a single PyTorch library, called Lightnet.

The main goal of this library was to create a modular set of building blocks, allowing you to easily integrate these models into various different applications. As such it is fairly bare bones and only provides a set of torch.nn.Modules, which a user can utilize to scaffold out an entire pipeline.
I took this approach even further by also constructing the various models out of different modules, allowing for rapid prototyping and development of new models.