Step1X-3D
收藏资源简介：
Step1X-3D 是一个开源框架，旨在解决3D生成中的数据稀缺、算法限制和生态系统碎片化等挑战。该框架通过一个严格的数据整理流程处理超过500万个资产，创建了一个具有标准化几何和纹理属性的高质量数据集。数据集的内容包括2M个高质量资产，其中大部分是从公开数据集中获取的。该数据集的创建过程包括多个阶段，包括对低质量纹理的消除、网格转换以确保几何一致性等。Step1X-3D旨在为可控3D资产生成领域的研究提供新的标准和开放性研究的新方向。
Step1X-3D is an open-source framework designed to address core challenges in 3D generation, including data scarcity, algorithmic limitations, and ecosystem fragmentation. Through a rigorous data curation pipeline, this framework processes over 5 million assets and constructs a high-quality dataset with standardized geometric and texture attributes. The dataset contains 2 million high-quality assets, most of which are sourced from public datasets. The creation process of this dataset involves multiple stages, such as eliminating low-quality textures and mesh conversion to ensure geometric consistency, among others. Step1X-3D aims to establish new benchmarks and open research directions for studies in the field of controllable 3D asset generation.
Step1X-3D 数据集概述
1. 数据集简介
Step1X-3D 是一个专注于生成高质量3D资产的开源框架，能够生成具有高保真几何和多样化纹理映射的3D资产，同时保持表面几何和纹理映射之间的卓越对齐。
2. 关键特性
- 高质量3D资产生成：包括基础几何（无纹理）、卡通风格、素描风格和照片级真实感的3D资产生成结果。
- 两阶段3D原生架构：结合了混合VAE-DiT几何生成器和基于SD-XL的纹理合成模块。
- 支持2D控制技术：如LoRA直接迁移到3D合成。
3. 模型下载
| 模型名称 | 下载链接 | 大小 | 更新日期 |
|---|---|---|---|
| Step1X-3D-geometry | Huggingface | 1.3B | 2025-05-13 |
| Step1X-3D-geometry-label | Huggingface | 1.3B | 2025-05-13 |
| Step1X-3D Texture | Huggingface | 3.5B | 2025-05-13 |
4. 开放数据集
| 数据来源 | 下载链接 | 大小 | 更新日期 |
|---|---|---|---|
| Objaverse | Huggingface | 320K | 2025-05-13 |
| Objaverse-XL | Huggingface | 480K | 2025-05-13 |
| Assets for texture synthesis | Huggingface | 30K | 2025-05-13 |
5. 依赖与安装
- Python 3.10
- CUDA 12.4
- PyTorch 2.5.1
- 其他依赖详见
requirements.txt
6. 许可证
Step1X-3D 采用 Apache License 2.0 许可证。
7. 引用
bibtex @article{li2025step1x3dhighfidelitycontrollablegeneration, title={Step1X-3D: Towards High-Fidelity and Controllable Generation of Textured 3D Assets}, author={Weiyu Li and Xuanyang Zhang and Zheng Sun and Di Qi and Hao Li and Wei Cheng and Weiwei Cai and Shihao Wu and Jiarui Liu and Zihao Wang and Xiao Chen and Feipeng Tian and Jianxiong Pan and Zeming Li and Gang Yu and Xiangyu Zhang and Daxin Jiang and Ping Tan}, journal={arXiv preprint arxiv:2505.07747} year={2025} }
- 1Step1X-3D: Towards High-Fidelity and Controllable Generation of Textured 3D AssetsStep1X-3D Team & LightIllusions Team StepFun & LightIllusions · 2025年
