With the rise of immersive media and its growing demand, researches related to multi-view images have been actively conducted. Identifying identical objects across different views is essential in this context, yet the inherent challenges of multi-view images, such as occlusion, make accurate object matching difficult. Moreover, the scarcity of publicly available multi-view datasets further impedes progress in this field. To address these challenges, we introduce a novel multi-view dataset with a variety of 3D scene characteristics, camera acquisition layouts, and an extensive range of objects and classes. We also propose a progressive matching algorithm, a three-stage process tailored to excel in scenarios plagued by occlusion. Finally, we improve the image segmentation results by refining the masks generated by existing networks with our matching results. Compared to existing instance matching algorithms, our method not only provides faster processing time but also significantly enhances performance, achieving an average ID matching accuracy of 97.5% and a mean Intersection over Union (mIoU) improvement of 19.8% on our dataset.

In the research of multi-view instance matching, the publicly available datasets are limited. To address this scarcity, we have developed a synthetic dataset, carefully designed to encapsulate various aspects of real-world multi-view scene acquisition scenarios. This dataset includes various kinds of factors such as the arrangement of cameras, variations in the number and class of objects within a scene, different viewing distances of each camera, and more. Consequently, our dataset spans a spectrum of complexity, ranging from relatively straightforward to highly challenging scenarios, pertinent to the task of multi-view instance matching. Henceforth, this dataset is referred to as the SMIIM (Synthetic Multi-view Images for Instance Matching) dataset.

The summary of SMIIM dataset is given in Table 1. Unreal Engine 5 was used to construct four distinct 3D scenes: Fruit, Animal, Airplane and People

Copyright (c) 2024 Intelligent Visual Media Laboratory (IVML)

All rights reserved.

The SMIIM dataset is copyrighted by the Intelligent Visual Media Laboratory (IVML). All rights are reserved. Unauthorized use, reproduction, or distribution of this dataset, or any portion thereof, is strictly prohibited without prior written permission from IVML. For any inquiries regarding the dataset, please contact us at hyung50300@hanyang.ac.kr. Permission is not granted, without written agreement and without license or royalty fees, to use, copy, modify, and distribute this dataset (the images and camera calibration data) and its documentation for any purpose.