in GitHub, Social media, Social network analysis

Gabriele Lami
Co-founder @ Elif Lab : www.eliflab.com

Darknet is a little awesome open source neural network written in C. Is fast, slim and friendly to use.

In particular, if you are interested in a fast and small classifier you should try Tiny Darknet. I’m using it in conjunction with a Bayesian network in order to classify a huge amount of images (using my slow pc) in a fast and reliable way (maybe I’ll explain more in another post).

You should alsouse Nightmare the counterpart of Deep Dream.

Like I did in a previous post, I wanted to analyze the community of projects surrounding Darknet on Github.

In this exploration I’m using the fork action in order to obtain a network. As second step I’m using a community detection method and some metrics in order to obtain interesting clusters.

Here are the results:

Community 1

Community 1 contains some of the reference projects regarding modern neural nets:

Keras, cntk, mxnet xgboost, caffe2.

It also contains the evergreens scikit-learn and opencv. The first is a pivotal node of machine learning in python, the second one is a masterpiece when talking about computer vision.

Interesting projects:

  • Pix2pix Image-to-image translation with conditional adversarial nets
  • Tiny-dnn header only, dependency-free deep learning framework in C++14

Community 2

This net is dominated by YOLO (You only look once) that is is a state-of-the-art, real-time object detection system.

Keras-yolo3, keras-yolo2,yolo2-pytorch, YOLO3–4-Py ,tensorflow-yolo ,caffe-yolo,YOLO_tensorflow, YOLO_Object_Detection

Interesting to read are also:

Community 3

Community 3 is centered on one of the biggest players when talking about deep learning: Tensorflow.

https://github.com/tensorflow/tensorflow

Tensorflow stands out in respect to the other players, at least in this community analysis.

Community 4

Mostly contains resources related to computer vision and deep learning.

Awesome projects included in this net are:

  • Dlib http://dlib.net/
  • Ncnn high-performance neural network inference framework optimized for the mobile platform by Tencent
  • TFFRCNN TFFRCNN FastER RCNN built on tensorflow

Community 5

This community contains some interesting little elements:

Community 6

Finally, this community shows our starting point, Darknet.

Since the starting point in this kind of analysis is (usually) the center of our net, the community here tends to be a bit generic.

Some interesting projects:

  • iGAN Interactive Image Generation via Generative Adversarial Networks
  • Insightface a face Recognition Project on MXNet

Interested in this or other data analysis techniques?
Need some help with your project? Contact us info@eliflab.com