Fashion IQ 2020 Challenge Dataset

该数据集为图像检索提供了一个独特的场景，其中输入的查询以候选图像和两个描述搜索目标视觉差异的自然语言表达式形式给出。该数据集包含了时尚图片以及与之相关的互动式图像检索的标题描述。评估参与者时，我们将使用召回率指标对测试集进行评估，具体包括在三个时尚类别（连衣裙、上衣、衬衫）中的Recall@10和Recall@50。数据集规模包括77,683张时尚图片和30,134对相对标题描述。任务是基于自然语言表达式描述的视觉差异进行图像检索。

This dataset offers a unique scenario for image retrieval, where the input query is structured as candidate images paired with two natural language expressions that describe the visual discrepancies between the search targets. This dataset includes fashion images and their corresponding caption descriptions for interactive image retrieval. When evaluating participants, we will assess the test set using recall metrics, specifically Recall@10 and Recall@50 across three fashion categories: dresses, tops, and shirts. The dataset comprises 77,683 fashion images and 30,134 pairs of comparative caption descriptions. The task is to conduct image retrieval based on visual differences described by natural language expressions.