top of page
3D GAN Pavilion
Filter by Industry Copy
Barcelona, Spain
Location:
2020
Year:
AI-Driven Pavilion Generation through Binary-Encoded 3D Models
This project pioneers a new approach to pavilion design by integrating Grasshopper and Generative Adversarial Networks (GANs) to create unique 3D structures through binary encoding. We trained GANs on a dataset of voxel-based pavilion models, each encoded as a sequence of 0s and 1s that represent positive and negative space within a bounding cube. This encoding technique transforms 3D models into simple 1-dimensional arrays, allowing for a streamlined representation of complex geometries.
To build the dataset, we developed a custom Grasshopper method that enables parametric control over pavilion features like pillar count, radius, positioning, canopy height, and mesh relaxation strength. The resulting models were generated in resolutions of 64x64x64, 128x128x128, and 256x256x256, with each voxel's inclusion determined by its binary value. The GAN training process, fed by these encoded variations, yields innovative pavilion designs that are both algorithmically diverse and architecturally coherent. This project demonstrates a novel design workflow, where AI and parametric tools collaborate to push the boundaries of generative architecture.
Team
Firas Safieddine
Michel Azzi
Yasser Sinjab
bottom of page