New Way to Generate 3D Meshes Using LLM

You name it, MeshGPT makes it.

ChatGPT can give you answers to almost any questions, but what if it could also make 3D meshes? Researchers from Technical University of Munich, Politecnico di Torino, and AUDI AG wondered the same and then presented MeshGPT, a new method of generating triangle meshes "that reflects the compactness typical of artist-created meshes."

We've seen AI tools that can generate 3D models, but what's interesting about MeshGPT is that it uses a large language model (LLM) to do this. 

Image credit: Yawar Siddiqui et al.

"We first learn a vocabulary of latent quantized embeddings, using graph convolutions, which inform these embeddings of the local mesh geometry and topology. These embeddings are sequenced and decoded into triangles by a decoder, ensuring that they can effectively reconstruct the mesh. A transformer is then trained on this learned vocabulary to predict the index of the next embedding given previous embeddings. Once trained, our model can be autoregressively sampled to generate new triangle meshes, directly generating compact meshes with sharp edges."

Image credit: Yawar Siddiqui et al.

While the vocabulary MeshGPT knows makes sure you get the object you asked for, this approach can also generate novel shapes, adding variety to the scene. If you give it an unfinished shape, it will suggest possible completions.

The researchers say that MeshGPT works better than state-of-the-art mesh generation methods, with a 9% increase in shape coverage. As a result, you get compact meshes with sharp geometric details for your 3D scene.

If you'd like to learn more about the method, check out the project's page here. Also, join our 80 Level Talent platform and our Telegram channel, follow us on InstagramTwitter, and LinkedIn, where we share breakdowns, the latest news, awesome artworks, and more.

Join discussion

Comments 0

    You might also like

    We need your consent

    We use cookies on this website to make your browsing experience better. By using the site you agree to our use of cookies.Learn more