Discrete Sampling

To explore the use of generative adversarial networks (GANs) in the design of 3-dimensional forms to probabilistically sample discrete architecture.
Jitesh Sapra & Matei Florescu

Workshop, 2020

The proposal aims to explore AI-driven architecture through the use of discrete architecture assemblies. It involves a bottom-up approach that leverages on the part to whole representation of discrete blocks and the high-dimensional representation of the GAN neural network to generate common objects such as chairs, cars etc.

3D Generative Adversarial Network
Derived from : Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling (Jiajun Wu*, Chengkai Zhang*, Tianfan Xue, William T. Freeman, and Joshua B. Tenenbaum)
Randomly Sampled Shapes (Chairs)
3D Generator Structure
Derived from : Learning a Probabilistic Latent Space of Object Shapes via 3D Generative-Adversarial Modeling (Jiajun Wu*, Chengkai Zhang*, Tianfan Xue, William T. Freeman, and Joshua B. Tenenbaum)
Random Sampling Trials
Interpolation between two samples
Interpolation iterations between samples
Design Exploration
The Inhuman Chairs
The concept for the proposal draws on the irony that a machine is meant to auto complete human tasks but in this context the proposal tries to complete the model of a chair designed by the AI bot. The design portrays the idea of the human parasitic invasion on the 3D GAN discrete chair model. The web like structures act as the tensile members providing structural stability to the chair by holding all the disjunct pieces of the chair together.

Base geometry of a chair generated through 3D Gan Sampling

Dense spider web-like tensile members wound around discrete samples
↑Back to Top