Image-to-Text Retrieval Using Object Detection

This study aims to enrich the image and use it more effectively through utilizing the image in different ways with object detection and thereby aims to improve the image-to-text retrieval studies. In this scope, a comprehensive analysis is presented using various methods on both monolingual and multilingual datasets. In addition to the traditional object detection architecture, it is aimed to obtain images that are more focused on objects by using zero-shot object detection and to compare it with the use of raw images. In this context, it is taken advantage of both the entire or part of the offline vocabulary and the open vocabulary. In addition, detected objects are also utilized as text and used to enrich the image representation. Thus, a comparative study is presented by exploiting the image with different methods through object detection.