Zhou, J., Li, X., Qi, L., & Yang, M.-H. (2024). LAYOUT-YOUR-3D: CONTROLLABLE AND PRECISE 3D GENERATION WITH 2D BLUEPRINT. arXiv. https://arxiv.org/abs/2410.15391
This paper introduces Layout-Your-3D, a novel method for generating complex 3D scenes from text prompts. The research aims to address the limitations of existing text-to-3D methods, which often struggle with generating plausible object interactions and lack user control over the generation process.
Layout-Your-3D utilizes a two-stage approach: a coarse 3D generation stage and a disentangled refinement stage. In the first stage, a 2D layout, either user-provided or LLM-generated, guides the creation of a reference image and individual 3D objects using efficient reconstruction models. The second stage refines the scene through a collision-aware layout optimization process, followed by instance-wise refinement to enhance individual object quality and enable customization.
The paper demonstrates that Layout-Your-3D outperforms existing text-to-3D generation methods in both qualitative and quantitative evaluations. The proposed method generates more plausible and visually appealing 3D scenes with significantly reduced generation time.
Layout-Your-3D presents a significant advancement in text-to-3D generation by enabling efficient and controllable creation of complex 3D scenes with plausible object interactions. The use of 2D layouts as blueprints and the two-step refinement process contribute to the method's superior performance.
This research holds significant implications for various applications, including virtual reality, robotics, and 3D content creation. The ability to generate high-quality 3D scenes from text prompts with user control opens up new possibilities for these fields.
While Layout-Your-3D demonstrates promising results, future research could explore extending the method to handle more complex scenes with a larger number of objects and diverse interactions. Additionally, investigating the integration of more sophisticated layout generation techniques could further enhance the controllability and realism of the generated 3D scenes.
A otro idioma
del contenido fuente
arxiv.org
Ideas clave extraídas de
by Junwei Zhou,... a las arxiv.org 10-22-2024
https://arxiv.org/pdf/2410.15391.pdfConsultas más profundas