标题： Unifi3D：在通用框架中生成和重建的3D表示研究 显示英文标题

标题： Unifi3D: A Study on 3D Representations for Generation and Reconstruction in a Common Framework

摘要： 随着文本和图像生成的快速进展，研究重点逐渐转向3D生成。 与图像中已建立的基于像素的表示不同，3D表示仍然多样且分散，涵盖了多种方法，如体素网格、神经辐射场、符号距离函数、点云或八叉树，每种方法都有其独特的优点和局限性。 在本工作中，我们提出一个统一的评估框架，旨在评估3D表示在重建和生成中的性能。 我们根据多个标准对这些表示进行比较：质量、计算效率和泛化性能。 除了标准模型基准测试外，我们的实验旨在得出3D生成流程中所有步骤的最佳实践，包括预处理、网格重建、使用自编码器的压缩以及生成。 我们的研究结果表明，重建误差显著影响整体性能，强调了需要联合评估生成和重建。 我们提供了可以指导选择适用于各种应用的3D模型的见解，有助于开发更强大和应用特定的3D生成解决方案。 我们框架的代码可在 https://github.com/isl-org/unifi3d 获取。 显示英文摘要

摘要： Following rapid advancements in text and image generation, research has increasingly shifted towards 3D generation. Unlike the well-established pixel-based representation in images, 3D representations remain diverse and fragmented, encompassing a wide variety of approaches such as voxel grids, neural radiance fields, signed distance functions, point clouds, or octrees, each offering distinct advantages and limitations. In this work, we present a unified evaluation framework designed to assess the performance of 3D representations in reconstruction and generation. We compare these representations based on multiple criteria: quality, computational efficiency, and generalization performance. Beyond standard model benchmarking, our experiments aim to derive best practices over all steps involved in the 3D generation pipeline, including preprocessing, mesh reconstruction, compression with autoencoders, and generation. Our findings highlight that reconstruction errors significantly impact overall performance, underscoring the need to evaluate generation and reconstruction jointly. We provide insights that can inform the selection of suitable 3D models for various applications, facilitating the development of more robust and application-specific solutions in 3D generation. The code for our framework is available at https://github.com/isl-org/unifi3d.