lightnet 3.1.0

Building blocks for recreating darknet networks in pytorch

Building blocks to recreate Darknet networks in Pytorch

Why another framework

pytorch-yolo2 is working perfectly fine, but does not easily allow a user to modify an existing network.
This is why I decided to create a library, that gives the user all the necessary building blocks, to recreate any darknet network. This library has everything you need to control your network, weight loading & saving, datasets, dataloaders and data augmentation.

Where it started as library to recreate the darknet networks in PyTorch, it has since grown into a more general purpose single-shot object detection library.

Installing

First install PyTorch and Torchvision.
Then clone this repository and run one of the following commands:

# If you just want to use Lightnet
pip install brambox   # Optional (needed for training)
pip install .

# If you want to develop Lightnet
pip install -r develop.txt

This project is python 3.6 and higher so on some systems you might want to use 'pip3.6' instead of 'pip'

How to use

Click Here for the API documentation and guides on how to use this library.
The examples folder contains code snippets to train and test networks with lightnet. For examples on how to implement your own networks, you can take a look at the files in lightnet/models.

If you are using a different version than the latest, you can generate the documentation yourself by running make clean html in the docs folder. This does require some dependencies, like Sphinx. The easiest way to install them is by using the -r develop.txt option when installing lightnet.

Cite

If you use Lightnet in your research, please cite it.

@misc{lightnet18,
  author = {Tanguy Ophoff},
  title = {Lightnet: Building Blocks to Recreate Darknet Networks in Pytorch},
  howpublished = {\url{https://gitlab.com/EAVISE/lightnet}},
  year = {2018}
}

Main Contributors

Here is a list of people that made noteworthy contributions and helped to get this project where it stands today!

• Tanguy Ophoff
• Jon Crall
• Cedric Gullentops