Comparative Analysis of Scene Generation Methods Based on Large Models
DOI:
https://doi.org/10.54097/s42mva48Keywords:
Large Models; Scenario Generation; Digital Testing; Multimodality; Model Evaluation.Abstract
Digital test scenario generation technology, a key support for intelligent algorithm verification and application, has broken through with the development of LLM. The paper identifies the core requirements of digital test scenario generation—including diversity and automation—then introduces the technical framework behind two key components: layout construction based on generative algorithms and intent reasoning powered by large models. It also conducts a comparative analysis of models such as Dungeon Alchemist and CityDreamer4D. To verify the effectiveness of these models, the paper employs both quantitative metrics and intelligent algorithm testing. Additionally, the paper points out existing shortcuts of the technology, such as its inadequate handling of complex scenario details, and offers insights into the generation of scenario’s future prospects, especially in enhancing reliability, adding efficiency, and advancing application in The field of digital twin technology. It also pays attention to personal local deployment and usability of the model. The study explores generative models that fit the needs of independent game developers and media producers, and outlines a vision for the development of such lightweight models moving forward.
Downloads
References
[1] Li X Y, Zhang Q, Kang D et al. Advances in 3D Generation A Survey. arXiv:2401.17807v1 [cs.CV] 31 Jan 2024.
[2] Liu, D Z. Liu Y, Huang W C, and Hu W. A Survey on Text-guided 3D Visual Grounding.arXiv:2406.05785v2 [cs.CV] 22 Jul 2024.
[3] Wen B C, Xie H Z, Chen Z X, Hong F Z, Liu Z W. 3D Scene Generation A Survey.arXiv:2505.05474v1 [cs.CV] 8 May 2025.
[4] Li H R, Tian Y R, Lan K, Liao Y. DreamScene 3D Gaussian-based Text-to-3D. arXiv:2507.13985v2 [cs.CV] 29 Jul 2025.
[5] Li H R, Shi H L, Zhang W L et al. DreamScene 3D Gaussian-based End-to-end Text-to-3D Scene Generation. arXiv:2507.13985. 2025.
[6] Radford A, Kim J W, Hallacy C, et al. Learning Transferable Visual Models From Natural Language Supervision. arXiv preprint arXiv:2103.00020, 2021.
[7] Gaidon A, Wang Q, Cabon Y, Vig E. Virtual Worlds as Proxy for Multi-Object Tracking Analysis. Dynamic Scene Generation and Evaluation. 2024.
[8] Xing T, Wu Y, Zhao Q C, et al. Digital Testing Scenario Generative Methods for Intelligent Algorithms Based on Large Language Models. Journal of Command and Control, 2025, 11(2): 239-247.
[9] Lv J X, Huang Y, Yan M F et al. GPT4Motion in Blender: Text-driven Dynamic Scene Generation with Physical Simulation. arXiv:2311.12631. 2024.
[10] Sun C Y, Han J L, Deng W J, Wang X L, Qin Z S, Stephen G. 3D-GPT: PROCEDURAL 3D MODELING WITH LARGE LANGUAGE MODELS. arXiv:2310.12945v2 [cs.CV] 29 May 2024.
[11] Zhou M, Hou J, Luo C, et al. SceneX: Procedural Controllable Large-scale Scene Generation via Large-language Models. arXiv preprint arXiv:2403.15698, 23 Mar 2024.
[12] Heusel M, Ramsauer H, Unterthiner T, et al. GANs Trained by a Two Time-Scale Update Rule Converge to a Local Nash Equilibrium//31st Conference on Neural Information Processing Systems. 2017.
Downloads
Published
Issue
Section
License
Copyright (c) 2025 Highlights in Science, Engineering and Technology
This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License.
