Relative geometry-aware siamese neural network for 6DOF camera relocalization

6DOF camera relocalization is an important component of autonomous driving and navigation. Deep learning has recently emerged as a promising technique to tackle this problem. In paper, we present novel relative geometry-aware Siamese neural network enhance the performance deep learning-based methods through explicitly exploiting geometry constraints between images. We perform multi-task predict absolute poses simultaneously. regularize shared-weight twin networks in both pose feature domains ensure that estimated are globally well locally correct. employ metric design adaptive distance loss learn capable distinguishing visually similar images from different locations. evaluate proposed method on public indoor outdoor benchmarks experimental results demonstrate our can significantly improve localization performance. Furthermore, extensive ablation evaluations conducted effectiveness terms function.