TripoSR: Fast 3D Object Reconstruction from a Single Image
Dmitry Tochilkin11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT David Pankratz11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT Zexiang Liu22{}^{2}start_FLOATSUPERSCRIPT 2 end_FLOATSUPERSCRIPT Zixuan Huang11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT Adam Letts11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT Yangguang Li22{}^{2}start_FLOATSUPERSCRIPT 2 end_FLOATSUPERSCRIPT Ding Liang22{}^{2}start_FLOATSUPERSCRIPT 2 end_FLOATSUPERSCRIPT Christian Laforte11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPT Varun Jampani1*1{}^{1*}start_FLOATSUPERSCRIPT 1 * end_FLOATSUPERSCRIPT Yan-Pei Cao2*2{}^{2*}start_FLOATSUPERSCRIPT 2 * end_FLOATSUPERSCRIPT 11{}^{1}start_FLOATSUPERSCRIPT 1 end_FLOATSUPERSCRIPTStability AI, 22{}^{2}start_FLOATSUPERSCRIPT 2 end_FLOATSUPERSCRIPTTripo AI
Abstract
This technical report introduces TripoSR, a 3D reconstruction model leveraging transformer architecture for fast feed-forward 3D generation, producing 3D mesh from a single image in under 0.5 seconds. Building upon the LRM [11] network architecture, TripoSR integrates substantial improvements in data processing, model design, and training techniques. Evaluations on public datasets show that TripoSR exhibits superior performance, both quantitatively and qualitatively, compared to other open-source alternatives. Released under the MIT license, TripoSR is intended to empower researchers, developers, and creatives with the latest advancements in 3D generative AI.
中文速览
从一张图片重建出三维物体，长期以来要么速度慢、要么质量差——TripoSR 正是为打破这一两难局面而生。它在大型重建模型 LRM 的 Transformer 架构基础上，通过精选训练数据、改进渲染方式、引入掩码损失（mask loss）和局部随机裁剪监督等多项技术，大幅提升了单图三维重建的质量与效率。在 GSO 和 OmniObject3D 两个公开数据集上的测试表明，TripoSR 在倒角距离（Chamfer Distance）和 F-score 两项指标上均超越 One-2-3-45、ZeroShape、TriplaneGaussian 等已有开源方法，同时在 A100 GPU 上仅需 0.5 秒即可生成带纹理的三维网格。该模型以 MIT 许可证完全开放，对希望探索三维生成式 AI 的研究者和开发者来说是一个即插即用的强力基线。
原文 arXiv:2403.02151;中英对照 + 大白话阅读 https://aha.fim.ai/paper/2403.02151v1