ashnkumar/sketch-code

Keras model to generate HTML code from hand-drawn website mockups. Implements an image captioning architecture to drawn source images.

What it solves

It converts hand-drawn website wireframes into working HTML code, automating a portion of the front-end development process by translating visual mockups into markup.

How it works

The project uses a deep learning architecture based on image captioning. It employs a CNN encoder to process the image and a LSTM decoder to generate the HTML as if it were a caption for the image. It was trained on a synthetic dataset of 1,700 images.

Who it’s for

Developers and designers who want to experiment with automated mockup-to-code translation using early deep learning techniques.

Highlights

  • Uses a CNN-LSTM architecture to treat HTML generation as an image captioning problem.
  • Supports batch processing of images to HTML conversion.
  • Includes scripts for training the model from scratch or fine-tuning pretrained weights.
  • Provides evaluation tools using the BLEU score to measure prediction accuracy.

Related

  • Project
  • Project
  • Project
  • Project
  • Project