This Form does not exist

Machine Learning, Design, DigitalFutures 2019

Intro

In order to find out a general way of translating 3D shapes into a language artificial neural network can understand, sculptures, especially those with human shapes, are chosen. Human shapes sculptures have similar ratio and styles, enabling machine learning technology to learn their styles instantly.

Role

Designer, Coder

location

Tongji University

Toolkits

Python, Keras, Tensorflow, Rhino, Adobe Photoshop, Adobe Illustratordcd

Duration

A week

GAN (Generative Adversarial Network)

machine learning and machine creation

GAN (Generative Adversarial Network) is implemented first, to learn the form of objects directly via pixelized 32×32 ×32 matrix.

ANN (Artificial Neural Network)

machine learning and machine creation

ANN, a simpler neural network built using Keras, is implemented , to learn the form of objects directly via contours of the forms.

Learning result is shown below.