Training StyleGAN with RunwayML #Training #GAN #MachineLearning #ArtificialIntelligence @RunwayML
RunwayML is a machine learning platform for creators. The platform boasts an impressive number of pre-trained algorithms in addition to some interactive demos. Several weeks ago they announced the Beta release of training on the platform! That said, training is currently limited to testing. If you would like to try it out and provide feedback, you can download RunwayML from their website and request training access through support. I went through that process this week and, once granted access, was able to get everything up and running in a couple of minutes.
RunwayML is currently using transfer learning on the StyleGAN model for training. StyleGAN is a Generative Adversarial Network that is able to create photorealistic images. The model was trained on thousands of images of faces from Flickr. RunwayML allows users to upload their own datasets and retrain StyleGAN in the likeness of your datasets.
…Transfer Learning leverages all of the image features it learned from the original data to speed up the process of learning features in your image dataset. Not only does this greatly reduce training time, it also reduces the number of images that you need to train your own model.
If you want to get started quickly you can also use one of RunwayML’s image datasets (mountains, forests, graffiti, etc). Selecting the smallest dataset (mountains) and training for two hours generated pretty passable landscape photos (below). If you’d like to learn more about training, check out the docs on the RunwayML website! If you’re interested in GANs or StyleGAN on RunwayML check out these videos.
Written by Rebecca Minich, Product Analyst, Data Science at Google. Opinions expressed are solely my own and do not express the views or opinions of my employer.
from Adafruit Industries – Makers, hackers, artists, designers and engineers! https://ift.tt/2E0XHHk
via IFTTT