Skip to content
This repository has been archived by the owner on Mar 7, 2024. It is now read-only.

Latest commit

 

History

History
39 lines (26 loc) · 1.83 KB

README.md

File metadata and controls

39 lines (26 loc) · 1.83 KB

Read Filtrou.me blog

https://filtrou.me/blog

How to build/deploy

  • '$ sh deploy.sh' on the root folder to build both sites (creator and player), put the built versions in the correct folder structure to be used with firebase hosting and deploy it.

  • 'npm run deploy' inside /functions/ to deploy firebase funtions.

  • You also need to set up security rules and cors rules for firebase hosting, storage and firestore.

Testing just the head tracker

  • go to /viewer/ and $ npm start

The notebook used to generate data and train the AI to solve PnP

  • /tf/all.ipynb (runs in ~15min on a Google Colab with GPU) Open In Colab

  • /tf/to_js.ipynb converts from Keras (.h5) to TF.JS

Project structure

  • /viewer/ is the visualizer, where all the 3D and AI code lives. It's a typescript project built with Parcel.
  • /creator/ is the creator of the filters. It uses /viewer/ inside an iframe. It's a typescript project built with Gatsby.
  • /functions/ are firebase functions. They're only used because I don't want to download the Firebase SDK and would rather make a simple HTTP request in the clients.
  • /tf/ contains the notebook used to generate fake data, train the model and then convert it to tf.js. It also has the trained model as a Keras export and tf.js export (quantized to 1 byte)

Acknowledgments and good links (these really helped me)