Object detection finds extensive application in the modern problem like surveillance,
autonomous vehicle, smart city application etc. In object detection, the objective is
to find the position and category of the object. For object detection deep learningbased
techniques like faster RCNN, YOLO series and SSD are developed. However,
it suffers from the domain shift problem. If the distribution between the source and
target domain data differs the performance of the algorithm degrades severely. To
circumvent this problem different domain adaptation technique are proposed. Among
these the source free domain adaptation (SFDA) is very practical in nature which
is based on the learned model trained with the source data and during the test of
the target data the source data is not available only the trained network is available.
Among this online domain adaptation-based which is a subset of SFDA based object
detection is very practical in nature. In this technique directly the target data is used
in an online manner for domain adaptation.