/LEGO-GAN

Generative Adversarial Networks (GANs) to trained to generate images of LEGO minifigures. Techniques adapted from the following research paper: https://arxiv.org/pdf/1511.06434.pdf

Primary LanguagePython

LEGO-GAN

A deep convolutional generative adversarial neural network to generate realistic images of LEGO minifigures

  • LEGO_DCGAN/minifig_images contains images of LEGO minifigures to train the model. Obtained via some web-scraping on Rebrickable.

  • LEGO_DCGAN/minifig_output contains generated images for each trial during training.

  • LEGO_DCGAN/generated: has several generated images by running LEGO_DCGAN/generated/inference.py. Give it a try yourself ;)

Below are some images generated by the DCGAN generator:

1. 64x64 (took about 7-8 hours of training on a CPU)

2. 128x128 (took about 10-12 hours of training on a CPU)

Learn more about implementing your own DCGAN in PyTorch here