tundra

Identification of people on videos based on their walking style using convolutional and recurrent neural networks.

To do

• Add a section about using your own neural network with tundra.

Installation

We use a lot of technologies, so you've got to do a lot of stuff.

You can do it step by step with this little tutorial or just scroll down to see a longer code snippet that you can just copypaste into your terminal.

Get required software first!

• Ruby and Ruby on Rails (Frontend Server)
  • You will have to run bundle install in front-server directory
• Node.js (Backend Prediction Server)
  • You will have to run npm install in prediction-server directory
• Torch7 (Neural Networks)
  • Torch7 installation guide
• ffmpeg (Data Preprocessing)
  • You download it here

Copypaste script for a quick start

git clone https://github.com/apaszke/tundra
cd tundra/front-server
bundle install
cd ../prediction-server
npm install

How to launch:

Run Frontend Server:

./run-frontend-server.sh

Run Prediction Server

./run-prediction-server.sh

How to use

Now you can open your browser at localhost:1337 and test the neural network.

Trivia

When it started

We've done a prototype of tundra during AGHacks 2015.

Inspiration

People can easily recognize the walking style of their friends. Let's see if we can teach a computer to identify walking people on videos.

What it does

It identifies people on videos based on their walking style.

How we built it

We build a neural network in torch and trained in on Amazon Web Service because we needed a lot of computing power. Additionally, we've made a website client in Ruby on Rails and a back-end server in Node.js which runs a trained neural network. When this neural network receives a video it returns the probabilities for every person it knows that the person is walking in the uploaded video.

Was it easy to create a neural network good enough for pitching at AGHacks 2015?

What's next for tundra-force dev team

We're going to...

Why 'tundra-force'

Well, the tundra word just sounds awesome, doesn't it?