Skip to content

redek91/ShapesAutoEncoder

Repository files navigation

ShapesAutoEncoder

An autoencoder that is capable to create squares using machinelearning.

test @ Netlify

Build, serve, train

Install needed packages with npm i

To generate Training data

  • Install processing and add processing-java.exe to PATH
  • Run the sketch from processing IDE or run the task run sketch in VS CODE (here is where the PATH is needed)

To train the model (GPU only)

Run npm run train

To try out

Run npm run start

To build

Run npm run build

Used frameworks

  • p5: for rendering
  • processing: for training data generation
  • tensorflow-js: for model training
  • webpack: .... just for testing it out...I guess

Todo-list

  • Create project structure
  • Create training data generator
  • Create auto encoder model and play with parameters (activation functions, loss, ecc.)
  • Create interface with p5
  • Add more shapes
  • Try Conv2d layers
  • Add rotation, translation to training data
  • Add colors to training data
  • Try out other datatypes (images??, sound??, gifs??)

Credits

This project is inspired by the Auto-Encoder-Demo project by Daniel Shiffman @CodingTrain GitHub - CodingTrain