Proposer une implémentation et validation d’un IA générative pour la création des objets 3D à partir des descriptions textuelles dans les modèles GAMA = Đề xuất một cách triển khai và kiểm định AI tạo sinh để tạo ra các đối tượng 3D từ các mô tả văn bản trong các mô hình GAMA

Abstract

This thesis delves into the application of generative artificial intelligence for creating 3D objects from textual descriptions within the framework of a serious game aimed at sustainable development education. Utilizing GAMA simulations and immersive visualizations in Unity suitable for virtual reality devices like the Meta Quest 3, our experimental approach involved generating 3D objects from textual prompts. These were rigorously evaluated using quantitative analyses based on CLIP models and qualitative evaluations through human perception. The findings demonstrate that integrating the MVDREAM diffusion model with the CRM reconstruction model maximizes efficiency with limited resources, significantly reducing the average generation time to 20 seconds per object under our resources. Additionally, this thesis outlines the development of an API that enhances accessibility to our 3D object generation model, tailored to be user-friendly and integrable into web services. Looking forward, we envision further embedding mesh simplification directly into our pipelines and broadening the reach of our technology to foster creativity and learning in diverse communities.