Darknet-NG  v1.0.0-1
next generation object detection using the darknet framework
Darknet-NG

Welcome to Darknet-NG! This is the next generation of fast + accurate object detection with Darknet/YOLO.

Note
This project is a work-in-progress. It does not build, and cannot be used! This page will be updated once we have a better idea of when the first release will be available.

Rationale

The Darknet/YOLO project is difficult to maintain. The original author – Joseph Redmon – is not involved in the development of Darknet/YOLO anymore. There have been hundreds of forks over the last few years, the most popular of which is the one by Alexey Bochkovskiy.

But this code is a convoluted mix of old-style C and C++, and is quite difficult to maintain and develop further. For this reason, the code was forked yet again in November 2022 to create Darknet-NG. This is a new codebase, rewritten to use C++17. These changes are an extensive overhaul and not intended to be merged back into the original codebase.

Goal

The initial goal is to have inference working with YOLOv4-tiny, YOLOv4-tiny-3L, and YOLOv4. After which we'll ensure that YOLOv7 and maybe the previous YOLOv3-tiny, YOLOv3-tiny-3L, and YOLOv3 configurations are also working. There are no plans at this time to look at older configurations such as YOLOv2.

Once inference is working, the next big feature will be to train new networks using Darknet-NG.

The configuration files and weights files will remain unchanged. People will have the opportunity to take their neural networks and easily switch between the "old" Darknet and the new Darknet-NG. This means the Darknet/YOLO neural networks will also continue working with OpenCV's DNN module and any other framework which already supports Darknet/YOLO.

Support

To help support this project:

Darknet-NG is written by St├ęphane Charette d.b.a. C Code Run
© 2022 St├ęphane Charette
http://www.ccoderun.ca/
steph.nosp@m.ane@.nosp@m.ccode.nosp@m.run..nosp@m.ca
steph.nosp@m.ane..nosp@m.chare.nosp@m.tte@.nosp@m.gmail.nosp@m..com
Tel. +1-250-769-2759   [Pacific Time, PST/PDT]