Updated Comparative Analysis between Recent 3D Object Generation Models. Adaptation to Immersive Technologies

Open Access
Article
Conference Proceedings
Authors: Geovana Amorim Abensur, Agustín A Ortiz Diaz, Sergio Cleger Tamayo, Delrick De Oliveira Freitas, Paulo Sergio Da Silva Farias, Gilberto Oliveira Neto, Gabriel Rossi Zanini
Abstract

Artificial Intelligence-based Content Generation (AIGC) and Extended Reality (XR) enable scalable creation of immersive 3D assets, but single-image-to-3D models still involve trade-offs between quality, responsiveness, and runtime performance. This study evaluates two generative models, Trellis and TripoSR, across Animals, Objects, Humanoids, and Places categories. The main comparison uses original generated meshes, while humanoid assets are additionally analysed after polygon reduction through original, intermediate, and low-polygon versions. Metrics are grouped into generation complexity, geometric accuracy, texture fidelity, perceptual realism and semantic consistency, and VR performance. Results show that Trellis achieves the best reconstruction quality but requires approximately 30 seconds for generation. TripoSR provides the fastest generation and stable XR performance, while Trellis reaches the highest CLIP-Similarity, indicating stronger semantic alignment but heavier meshes. These findings show that model selection for XR depends on the target application, balancing reconstruction quality, semantic consistency, and runtime constraints.

Keywords: Artificial Intelligence Generated Content, 3D Generative Content, Extended Reality, Quantitative Metrics, 3D Real-Time Rendering

DOI: 10.54941/ahfe1008249

Cite this paper
Downloads
0
Visits
1
Download PDF

More from this volume

← Physiological Indicators of Trust in AI-Assisted Air Traffic Control: A Multimodal Eye-Tracking and EEG StudyEvaluation of Gesture Separability and Stability Based on Surface Electromyography for General Interaction Design →
View all articles in Human Factors in Design, Engineering, and Computing →