4D-DRESS - 包含语义注释的真实世界4D服装数据集
收藏资源简介:
4D-DRESS是由苏黎世联邦理工学院和图宾根大学联合构建的首个真实世界的4D服装数据集。该数据集包含64套不同人物服装的520多个动作序列,总计78000个纹理扫描,涵盖了丰富的服装类型和动态动作。4D-DRESS通过半自动的4D人体解析流程,结合人工干预与自动化技术,精确标注了服装与人体动作的复杂细节,为服装模拟和重建等多个计算视觉和图形学领域的任务设立了新基准。该数据集的发布旨在缩小现有服装算法与现实世界人类服装之间的差距,促进数字虚拟角色服装研究的发展,为研究人员提供了宝贵的数据资源。
4D-DRESS is the first real-world 4D clothing dataset jointly constructed by ETH Zurich and the University of Tübingen. This dataset includes over 520 motion sequences of 64 different human outfits, totaling 78,000 texture scans, covering a wide range of clothing types and dynamic movements. Through a semi-automated 4D human parsing process, combining manual intervention with automated technology, 4D-DRESS precisely annotates the complex details of clothing and human movements, setting a new benchmark for tasks in computational vision and graphics such as clothing simulation and reconstruction. The release of this dataset aims to bridge the gap between existing clothing algorithms and real-world human attire, promoting the development of digital virtual character clothing research and providing researchers with valuable data resources.
数据集概述
4D-DRESS 是一个专注于真实世界人类服装的4D数据集,包含64种不同的人类服装,涵盖超过520个运动序列和78,000帧扫描。每个运动序列包含以下内容:
- 高质量的4D纹理扫描。
- 精确的顶点级语义标签。
- 对应的提取衣物网格和拟合的SMPL(-X)身体网格。
数据集详细记录了动态运动中的4种连衣裙、28种下装、30种上装和32种外衣。每种衣物还提供了其规范模板网格,以支持未来的人类服装研究。
数据集内容
- 4D-DRESS Dataset:包含多种人类服装的4D扫描数据。
- 4D-Human-Parsing Code:用于解析4D扫描数据的代码。
数据集结构
数据集的文件结构如下:
4D-DRESS └── < Subject ID > (00***) └── < Outfit > (Inner, Outer) └── < Sequence ID > (Take*) ├── basic_info.pkl ├── Meshes_pkl ├── SMPL ├── SMPLX ├── Semantic ├── Capture └── Useful tools for 4D-DRESS
数据集使用工具
- aitviewer:用于可视化4D-DRESS序列的工具。
- extract_garment.py:用于提取标记的衣物网格和渲染多视角像素标签的工具。
4D 人类解析方法
该方法首先将当前和前一帧扫描渲染成多视角图像和标签,然后收集多视角解析结果,并通过图割算法优化顶点标签。用户可以通过检查多视角渲染标签轻松引入手动校正标签。
数据集引用格式
@inproceedings{wang20244ddress, title={4D-DRESS: A 4D Dataset of Real-world Human Clothing with Semantic Annotations}, author={Wang, Wenbo and Ho, Hsuan-I and Guo, Chen and Rong, Boxiang and Grigorev, Artur and Song, Jie and Zarate, Juan Jose and Hilliges, Otmar}, booktitle={Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR)}, year={2024} }




