Sketch-Based Texture Generation Using a Neural Network

Spring 2019

https://github.com/e2-dunstan/AT-DeepLearning

  • Python with Tensorflow
  • Around 800 training images
  • Convolutional neural network
  • 40 hours to train 1000 epochs
  • Generates images at runtime
  • Can easily determine windows and building boundaries
  • Behaviour to create outlined images from a source image for the training data manually programmed
  • Enlarge image
  • Enlarge image

Advanced Technologies Showreel, featuring this project.

Play Video

Report

  • 2 May 2019
  • Sketch-Based Texture Generation Using a Neural Network Technical Report
  • The technical report written for my advanced technologies module for sketch-b...