Domain adaptation of object detector using scissor-like networks

Lin Xiong,Mao Ye,Dan Zhang,Yan Gan,Dongde Hou

DOI: https://doi.org/10.1016/j.neucom.2021.05.012

IF: 6

2021-09-01

Neurocomputing

Abstract:<p>When the training data and the test data do not obey the same distribution, the performances of many object detection methods always decrease greatly. Naturally, domain adaptation methods at feature level are proposed. The basic idea is to adapt the feature extraction network such that the feature distributions of the source and target domains match. We propose a new method that is built directly on the Faster R-CNN model, which not only aligns the source and target data features, but also forces their generated features closer together to further align the source and target domains. Moreover, compared with previous approaches, we construct a more powerful discriminator and a simple generator to solve the domain adaptation problem. The model works like a pair of scissors, so we call it Scissors Networks (SN). We conduct extensive experiments on popular datasets, including Cityscapes, Foggy Cityscapes, SIM10k and KITTI. The experimental results demonstrate that our algorithm is superior to the state-of-the-art deep learning based domain adaptation approaches.</p>

computer science, artificial intelligence

What problem does this paper attempt to address?

The problem that this paper attempts to solve is that in object detection tasks, when the training data and the test data do not follow the same distribution, the performance of many object detection methods will drop significantly. Specifically, when the data in the target domain has no labels or only a small number of labels, how to make the object detector trained on the source domain also perform well on the target domain. To solve this problem, the paper proposes a new method based on the Faster R - CNN model. This method not only aligns the data features of the source domain and the target domain, but also further makes the generated features closer through the Generative Adversarial Networks (GANs), thereby strengthening the alignment between the source domain and the target domain. In addition, compared with previous methods, this method constructs a more powerful discriminator and a simple generator to solve the domain adaptation problem. The entire model works like scissors, so it is called Scissors Networks (SN). Through extensive experiments on multiple popular datasets, it is proved that this algorithm is superior to the existing deep - learning - based domain adaptation methods.