Object detection is one of the most important challenges in computer vision. Object detectors are usually trained on bounding-boxes from still images. Recently, video has been used as an alternative source of data for learning an object detector. Yet, for a given target test domain (image or video), the performance of the detector depends on the domain it was trained on. In this paper, we examine the reasons behind this performance gap. We define and evaluate different domain shift factors: spatial location accuracy, appearance diversity and image quality. We examine the impact of these factors by comparing the performance of object detectors before and after factoring them out. The results show that all three factors affect the performance of the detector and are able to explain a significant part of the performance gap. However, even after factoring them out, we still observe that training on images outperforms training on video frames when testing on still images and vice versa. This indicates that there exist domain specific correlation between training and test set, most likely due to the similarity of objects within a domain, which we investigate and explain.

本文研究对象检测在图像与视频领域中的性能差异，探究其原因并分别评估了空间位置准确性、外观多样性、图像质量和比例分布等四种不同领域变量的影响，发现这些因素对检测器的性能都有影响，并且综合效应可解释近乎全部性能差异。

分析视频和图像之间目标检测领域漂移因素