Apparel Generation via Cluster-Indexed Global and Local Feature Representations
Apparel Generation via Cluster-Indexed Global and Local Feature Representations
复制标题
DOI：
10.1109/gcce50665.2020.9291984
复制
发表时间：
2020-10
期刊：
影响因子：
--
通讯作者：
Chunzhi Gu;Zhengyu Huang;Sicheng Li;Haoran Xie;Xi Yang;Chao Zhang
中科院分区：
文献类型：
--
作者：
Chunzhi Gu;Zhengyu Huang;Sicheng Li;Haoran Xie;Xi Yang;Chao Zhang
服装设计需要美学方面的专业知识，这对于非专业的普通用户来说是具有挑战性的。受数据科学最新进展的启发，在本文中，我们通过利用深度神经网络模型以简单的方式解决服装生成任务。我们建议通过三个选择步骤从大图片中生成衣服（例如，衣服的类型）到细节（例如，颜色）通过改变集群ID和潜在变量。用户可以通过这些步骤来实现理想的设计。在公开数据集上的实验证明了该方法的有效性。
Apparel design requires expertise in aesthetics, which is challenging for non-professional general users. Inspired by the recent advances in data science, in this paper we address the task of apparel generation in a simple way by leveraging a deep neural network model. We propose to generate clothes through three selection steps from the big picture (e.g., type of clothes) to the details (e.g., color) by varying the cluster ID and latent variables. Users can go through these steps to achieve an ideal design. Experiments on a publicly available dateset demonstrate the effectiveness of our method.