terrorflow-hackathon
TerrorFlow, 24 hour AIVD hackathon project
Video
Description
This repository contains the code for the AIVD (Dutch National Intelligence and Security Agency) hackathon. We created a system for automatic detection of new extremists which may be worth keeping a closer eye on for the AIVD.
We do this by recognizing certain aspects of images, of which the combination may be especially interesting. For instance, an AK-47 in an image on Instagram is not that interesting, but an AK-47 in a desert environment with a Dutch burning flag may be very interesting.
This is a multi-label computer vision problem which we tackled with deep learning (convolutional neural networks).
The scope of this project was pretty large. Aside from creating the models, we also created a flashy browser front-end and a back-end and deployed it on an 8 GPU AWS instance. This was all created in 24 hours, so don't expect the code to be of high quality.
Team
Our team consisted of four people with a machine learning/data science background.
- Bas van Berkel @basvb
- Joris van Vugt @jvanvugt
- Steven Reitsma @StevenReitsma
- Guido Zuidhof @gzuidhof
Tech
- TensorFlow
- Keras
- Vue.js
- Flask
- AWS