Let's Go Shopping (LGS)

Let's Go Shopping (LGS) 数据集是一个大规模的公开数据集，包含从公开的电子商务网站收集的1500万对图像-标题数据。该数据集专注于前景对象，背景较为简单，适用于视觉概念理解。LGS数据集的图像质量高，专注于电子商务领域，适用于视觉-语言双模态任务，如图像标题生成和文本到图像的生成，能够帮助模型生成更丰富的标题并实现电子商务风格转换。The Let's Go Shopping (LGS) dataset is a large-scale publicly available dataset containing 15 million image-caption pairs collected from public e-commerce websites. This dataset focuses on foreground objects with relatively uncluttered backgrounds, making it suitable for visual concept understanding tasks. The LGS dataset features high-quality images and is specialized in the e-commerce domain, applicable to vision-language bimodal tasks such as image captioning and text-to-image generation, which can help models generate more diverse captions and achieve e-commerce style transfer.

搜集汇总

背景与挑战

背景概述

Let's Go Shopping (LGS)是一个包含1500万图像-文本对的大规模数据集，源自电子商务网站，特点是图像聚焦前景对象且背景简单。该数据集适用于视觉概念理解和视觉-语言双模态任务，能提升图像描述生成和文本到图像生成的性能。