ST-Net-dataset
收藏资源简介:
ST-Net-dataset是由哈尔滨工业大学(深圳)研究团队构建的大规模数据集,专门用于无监督的协同服装合成任务。该数据集包含30,876张上下装服装图片,涵盖了多种风格和纹理的服装。数据集的创建过程通过图论方法和感知图像块相似性(LPIPS)确保训练集和测试集之间没有重叠。数据集的应用领域主要集中在时尚设计和服装生成,旨在通过自监督学习生成与给定服装风格和纹理相匹配的协同服装,解决传统方法依赖配对数据集的问题。
ST-Net-dataset is a large-scale dataset developed by the research team at Harbin Institute of Technology (Shenzhen), exclusively tailored for unsupervised collaborative clothing synthesis tasks. It comprises 30,876 images of upper and lower garments, covering diverse clothing styles and textures. During the dataset construction, graph theory approaches and Learned Perceptual Image Patch Similarity (LPIPS) are utilized to ensure no data overlap between the training set and the test set. The primary application scenarios of this dataset lie in fashion design and clothing generation, where it aims to generate collaborative garments that match the style and texture of a given clothing item via self-supervised learning, thereby resolving the limitation of traditional methods that rely on paired datasets.
数据集概述
数据集名称
ST-Net-dataset
数据集任务类别
- 图像到图像(image-to-image)
数据集规模
- 10K < n < 100K(包含30,876张图像)
数据集用途
该数据集用于自监督的搭配服装合成(Collocated Clothing Synthesis, CCS)任务,涵盖上衣和下装两类服装。
数据集划分
- 数据集被划分为训练集和测试集,比例为4:1。
- 划分适用于“上衣 → 下装”和“下装 → 上衣”两种设置。
- 通过图基方法和学习感知图像块相似性(LPIPS)确保训练集和测试集之间没有重叠。
图像分辨率
- 所有图像的分辨率为256 x 256像素。
引用
如果使用该数据集,请引用以下论文:
@inproceedings{dong2024towards, title={Towards Intelligent Design: A Self-Driven Framework for Collocated Clothing Synthesis Leveraging Fashion Styles and Textures}, author={Dong, Minglong and Zhou, Dongliang and Ma, Jianghong and Zhang, Haijun}, booktitle={ICASSP 2024-2024 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, pages={3725--3729}, year={2024}, organization={IEEE} }

- 1Towards Intelligent Design: A Self-driven Framework for Collocated Clothing Synthesis Leveraging Fashion Styles and Textures哈尔滨工业大学(深圳) · 2025年



