/Pix2code-Screenshot-to-code-dataset-builder

Dataset builder toolkit for Pix2code Screenshot-to-code

Primary LanguagePython

Overview

This is the custom dataset builder, web-dsl-mapping and bootstrap.vocab generator for Screenshot-2-Code and Pix2code.

It's based off of the Pix2code library by Tony Tony Beltramelli, but in particular the screenshot-2-code version by Emil Walner which uses Keras + Tensorflow.

It allows you to create your own custom datasets as well as vocab. It also includes training scripts for AWS, Colab, and local cpu. Folders:

  • /dataset-builder
    • Build your dataset here, contains your .npz and .gui pairs for model training.
  • /dsl-builder
    • This is where you build your dsl library (web-dsl-mapping.json + bootstrap.vocab).
  • /training
    • Contains training scripts for Colab, AWS and local for model training

My build: https://www.youtube.com/embed/Jr6DyjczXUA (Note the weights file is not included in this project).