Machine learning database generation (Personal prototype)
Concept for a procedural 3D object / room generator. This first example generates different models of chairs using a custom library of 3D parts, and generates random textures and shaders using Substance Designer as a base.
The lighting is also randomized.
This prototype can export thousands of different chair types using an infinity of lighting and camera situations.
It exports several passes including a mask pass, a metallic pass, scene normal, albedo, etc.
Final output is a PNG collection.
Made using visual scripting and minimal C++.
Thousands of images can be created in a matter of hours on a mid range consumer GPU using an SSD.
A couple of selected examples below with various passes. Each image also contains metadata with lighting and camera information.

You may also like

Back to Top