TMBuD-v2: A Multi-Purpose Urban Landmark Detection Dataset

Recognition and identification of urban landmarks is not a facile task and naturally blends several research topics like image signal processing, computer vision, augmented reality. In this paper we introduce TMBuD-v2 - Timişoara Building Dataset - a multi-purpose dataset for urban landmark recognition. The TMBuD-v2 building dataset contains 1360 images with the resolution of 768 × 1024 pixels taken using a mobile phone from a street view perspective. The images are organized into 144 district landmarks located in several tourist parts of Timisşoara. The proposed dataset offers the possibility of evaluating landmark recognition solutions in urban scenario, salient edges on the buildings, and semantic segmentation of the urban environment. The dataset is available at https://github.com/CipiOrhei/TMBuD

I. Introduction

Urban landmarks can be defined from two perspectives: one as a structure that is easy to identify, and the second as a building or place that has an important historical importance. Landmarks serve as “spatial magnet” in which cultural, civic, or economical activities concur. In this sense, they have become an important aspect in multiple domains related to tourism and culture. In the modern era, the interplay of multi-media technologies with urban icons not only enriches public engagement but also sparks scholarly intrigue, highlighting their multifaceted significance [1], [2].