07/29/2026
Congratulations to the researchers who recently published exciting new research for the field of AI-generated 3D models in the journal Transactions on Machine Learning, or TMLR! The team includes three faculty from The GAME School, a current ASU computer science PhD student and two ASU computer science alumni in partnership with two private sector collaborators.
Their paper, titled “DecompDreamer: A Composition-Aware Curriculum for Structured 3D Asset Generation,” presents a new AI technique that helps computers create more realistic 3D scenes from text descriptions by prioritizing optimization schedules.
While existing AI tools can generate individual objects well, they often struggle with more complex prompts involving multiple objects and their relationships. The new method, called DecompDreamer, solves this by first arranging the overall scene and then refining each object one at a time. The result is more accurate, detailed, and believable 3D content, with potential applications in game development, virtual reality, digital design and animation.
Their research was also awarded the selective J2C Certification, which designates the paper as being on the joint NeurIPS/ICLR/ICML journal-to-conference track. Receiving this certification gives the researchers the opportunity to present their findings at one of these three prestigious conferences.
Read more! 👉 https://news.asu.edu/b/20260728-asu-researchers-privatesector-collaborators-publish-research-aigenerated-3d-models
Authors: Utkarsh Nath, Rajeev Goel, Rahul Khurana, Kyle Min, Mark Ollila, Pavan K. Turaga, Varun Jampani, Tejaswi Gowda
You can read the full paper here: https://openreview.net/pdf?id=3qy4J6QFbn