2015/01/06 by Vicky Kalogeiton, Kalogeiton, Vicky, Vittorio Ferrari +3
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #Domain Adaptation and Few-Shot Learning #FOS: Computer and information sciences #Infrared Target Detection Methodologies
paper · pdf · doi:10.48550/arxiv.1501.01186
openalex publication_date 2015/01/06 · openalex created_date 2022/10/03 · openalex updated_date 2026/07/28
Object detection is one of the most important challenges in computer vision.\nObject detectors are usually trained on bounding-boxes from still images.\nRecently, video has been used as an alternative source of data. Yet, for a\ngiven test domain (image or video), the performance of the detector depends on\nthe domain it was trained on. In this paper, we examine the reasons behind this\nperformance gap. We define and evaluate different domain shift factors: spatial\nlocation accuracy, appearance diversity, image quality and aspect distribution.\nWe examine the impact of these factors by comparing performance before and\nafter factoring them out. The results show that all four factors affect the\nperformance of the detectors and their combined effect explains nearly the\nwhole performance gap.\n