BANG: Dividing 3D Assets via Generative Exploded Dynamics

Summary

This paper talks about BANG, a new AI system that can take a 3D object and break it down into its individual parts in a smooth and controlled way, similar to how an exploded diagram shows all the pieces of a machine separated but connected.

What's the problem?

The problem is that current 3D design tools require a lot of manual work and artistic skill to divide complex objects into parts, which is slow and difficult. This makes creating and understanding detailed 3D models challenging.

What's the solution?

BANG uses a special generative model that learns from many examples how to gradually explode and separate an object into parts while keeping the shape and meaning of each part intact. It allows precise control over which parts to separate and how, and it can even interact with other AI models like GPT-4 for more intuitive design through 2D images or sketches.

Why it matters?

This matters because it makes 3D modeling easier, faster, and more creative, helping artists, designers, and engineers build detailed and functional 3D models for things like video games, virtual reality, 3D printing, and manufacturing.