October 23, 2025 | Duke I3T Paper Evaluating 3D Object Generation Models Wins the Best Paper Award at a Workshop Co-Located with IEEE ISMAR
In a paper led by Duke I3T PhD student Yanming Xiu and coauthored by two of lab's high school affiliates, Joshua Chilukuri and Shunav Sen, who Yanming is mentoring via the North Carolina School of Science and Mathematics (NCSSM) Mentorship Program, we evaluated four different speech-to-3D object generation pipelines for augmented reality applications, including via an IRB-approved user study that assessed the generated content's realism, texture quality, mesh integrity, whether the content reflects the user’s intent, and whether the users are annoyed by the latency the speech-to-3D object generation pipelines require.
Our study demonstrates that modern 3D generation pipelines, when integrated into AR systems, are capable of producing assets that rival or even surpass moderately detailed handcrafted models in perceptual quality while offering a nearly real-time feedback experience to users. This highlights the growing potential of generative AI to assist in rapid AR content creation, especially for developers and creators without access to professional asset design resources.
This paper received the IEEE Universal Augmented Interaction Workshop Best Paper Award.
