CSTBIR (Composite Sketch+Text Based Image Retrieval)
收藏资源简介:
CSTBIR数据集是由印度理工学院贾伊普尔分校的研究人员创建的,包含约200万个查询和108000个自然场景图像。该数据集由手绘草图、补充的自然语言文本描述和相关自然场景图像组成,旨在解决非母语人士在搜索难以命名的特定对象时的问题。数据集中的图像和文本描述来源于Visual Genome和Quick, Draw!数据集,经过处理后形成独特的复合查询,适用于对象检测、分类和检索等任务。
The CSTBIR dataset was created by researchers from the Indian Institute of Technology (IIT) Jaipur. It contains approximately 2 million queries and 108,000 natural scene images. The dataset is composed of hand-drawn sketches, supplementary natural language text descriptions and associated natural scene images, aiming to address the challenges faced by non-native speakers when searching for specific objects that are difficult to name. The images and text descriptions in the dataset are sourced from the Visual Genome and Quick, Draw! datasets, and have been processed into unique composite queries applicable to tasks such as object detection, classification and retrieval.




