PROJECT TITLE :

Class Agnostic Image Common Object Detection

ABSTRACT:

In computer vision, determining the similarity between two images is a major challenge. Existing work on picture similarity generation mostly focuses on global feature distance calculation, local matching of features and comparison of image ideas. Class-agnostic common items from two photos have not before been explored, which goes a step farther in terms of capturing image similarities at the region level. End-to-end CODN is proposed in this paper to find class-agnostic common items between two photos. Locating and matching are the two primary components of the suggested technique. Candidates for each two photos are generated by the locating module. Pairs of proposals are matched using two photographs, and their bounding boxes are refined based on this information. CODN's learning technique is integrated, and a multi-task loss is devised to ensure that both region localization and common object matching are ensured by the implementation of the multi-task loss. PASCAL VOC 2007 and COCO 2014 datasets are used in the experiments. The outcomes of the experiments show that the recommended method works.