Zero-Shot Object Detection and Segmentation: A Focus on Street View Imagery

Abstract

With the advancement of data collection technologies, the importance of new data types like street view images, in addition to satellite and aerial images, has increased. Street view images (SVI) stand out by containing more comprehensive and real-time information compared to other types of images, thus offering a rich research field for object detection and segmentation processes. Interpreting and analyzing complex street view images requires accurate and effective processing of these data types. The use of SAM (Segment-Anything Model) and Grounding DINO models, which are less emphasized in the literature on street view images, forms the focus of this study. The application of these two models provides the opportunity to successfully perform segmentation and detection processes together on street images. This approach marks a significant advancement in the analysis of street images within the field of visual data processing, enhancing efficiency.