Face2Text

Face2Text数据集由马耳他大学的RIVAL研究组创建，旨在收集人类面部图像的丰富文本描述。该数据集包含400张从野外拍摄的公众人物面部图像，每张图像至少有3个描述，总计收集了1400个描述。创建过程中，通过定制的众包网站收集描述，并进行了初步的标注研究。数据集主要应用于自动面部描述生成和识别技术，旨在推动图像到文本技术的发展，特别是在面部描述的精细化和属性聚焦方面。

The Face2Text dataset was developed by the RIVAL Research Group at the University of Malta, with the objective of collecting rich textual descriptions of human facial images. This dataset contains 400 facial images of public figures captured in the wild, with each image paired with a minimum of three descriptions, leading to a total of 1400 collected descriptions. During the dataset construction phase, descriptions were collected via a custom-built crowdsourcing platform, and preliminary annotation studies were performed. The dataset is primarily applied to automatic facial description generation and recognition technologies, and aims to advance the development of image-to-text technologies, especially in the fine-grained and attribute-focused domains of facial description.