DigitalFUTURES 2021
Machine Intelligence in Architecture 2.0
Harvard Graduate School of Design

RunwayML – StyleGAN

Part 01 of the workshop focused on learning how to use RunwayML for image generation and style transfer. An intuitive way to use machine learning tools for media generation.

For this exercise, I chose Robgonsalves’s impressionist landscapes paintings dataset (5000 images) for training StyleGAN.

Download the dataset from here.

Process:
Data Collection – Pre-trained model (Landscapes) – Training – Generating outputs from the latent space

2.5D Latent Walk

Part 02 of the workshop focused on connecting StyleGAN to Rhino-Grasshopper for real-time inference and to generate 2.5D/3D shapes using image processing and image lofting techniques.