From Flat to Spatial - Comparison of 4 methods constructing 3D, 21/2D Models from 2D Plans with neural networks

Oct 24, 2026

In architecture, converting single images into 2½D and 3D meshes improves visualization quality and production efficiency. This paper compares four methods—“One-2-3-45,” “CRM: Single Image to 3D Textured Mesh with Convolutional Reconstruction Model,” “Instant Mesh,” and “Image-to-Mesh”—evaluating their suitability for design visualization from minimal inputs such as photographs or sketches. The study emphasizes faster prototyping, higher reconstruction fidelity, and more accurate texture generation to support flexible architectural iteration and client communication.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages12

Reading Time30 min

AccessFree to download

Abstract

Keywords

Introduction

From Flat to Spatial: Comparison of 4 methods constructing 3D, 21/2D Models from 2D Plans with neural networks. Jacob Sam 1, Karan Patel2, Mike Saad3 Abstract In the field of architecture, the conversion of single images into 2½D and 3D meshes is a promising technology that enhances design visualization and efficiency. This paper evaluates four innovative methods: "One-2-3-45," "CRM: Single Image to 3D Textured Mesh with Convolutional Reconstruction Model," "Instant Mesh," and "Image-to-Mesh." These methods are at the forefront of this technology, focusing on their applicability in architectural design and visualization. They streamline the creation of 3D architectural models, enabling rapid prototyping and detailed visualization from minimal initial inputs, such as photographs or simple sketches.One-2-3-45 leverages a diffusion-based approach to generate multi-view reconstructions, ensuring high geometric fidelity and texture quality. CRM utilizes a convolutional network to integrate geometric priors into its architecture, producing detailed and textured meshes quickly and efficiently. Instant Mesh combines the strengths of multi-view diffusion and sparse-view models to offer speed and scalability, suitable for diverse architectural projects. Image-to-Mesh leverages a generative adversarial network (GAN) to produce 3D meshes from single images, focusing on maintaining high texture fidelity and geometric accuracy by incorporating image and depth map data into its training process. It uses a hybrid approach that combines voxel-based representations with surface reconstruction techniques to ensure detailed and realistic 3D models.This comparative study highlights each method's contribution to reducing design cycle times, improving accuracy, and enabling flexible adaptations to various architectural styles and requirements. By providing architects with powerful tools for rapid visualization and iteration, these advancements in 3D mesh generation are set to revolutionize architectural practices. Keywords: Applied Research, scale XL, Image to 3d, Mesh Reconstruction, Generative Design Introduction The evolution of digital tools within the architectural domain has significantly reshaped how architects engage with the design and visualization of structures. The capacity to translate twodimensional imagery into 21/2d,3d models is not merely a technical advancement but a paradigm shift that augments the creative process, enhances client communication and refines the realization of architectural concepts. As the demand for faster and more accurate visual representations grows, the development of methods that facilitate the quick conversion of single images into detailed 3D meshes is becoming increasingly vital. These technologies are crucial in environments where rapid iteration and visualization are required, from preliminary design phases to the final presentation stages. (Yi et al. 2023)(2024)(Sun et al. 2018) Architectural visualization has traditionally relied on physical models and CAD drawings, which can be time-consuming and resource-intensive. However, recent advancements in single-image 3D mesh generation technologies offer a more efficient way to create detailed 3D models (Kanazawa et al., 2018) . These technologies enable architects to quickly iterate and explore various design options without the usual overhead associated with 3D modeling (Song et al., 2017). By using single images, these methods can automatically estimate 3D shapes, cameras, and textures of objects, significantly reducing the manual effort required (Pan et al., 2019) . The emergence of deep learning methods has further enhanced the capabilities of 3D mesh generation. Techniques like "Deep Mesh Reconstruction From Single RGB Images via Topology Modification Networks" have shown superior performance in generating 3D shapes with complex topologies (Pan et al., 2019) . Additionally, approaches such as "Pixel2Mesh: Generating 3D Mesh Models

FAQs

What problem does the paper address in architectural design workflows?

It addresses the need to convert single images into 2½D and 3D meshes to speed up visualization and reduce the manual effort of creating architectural models.

Which four 3D mesh generation methods are compared?

The paper compares “One-2-3-45,” “CRM: Single Image to 3D Textured Mesh with Convolutional Reconstruction Model,” “Instant Mesh,” and “Image-to-Mesh.”

What is the main outcome the comparison is intended to support?

The comparison aims to show how these methods reduce turnaround times, improve geometric accuracy, and provide tools that adapt to different architectural styles and project requirements.

Oct 24, 2026

Upload by RuangKosong