This paper presents a neural network-based Unscented Kalman Filter (UKF) to track the pose (i.e., position and orientation) of a known, noncooperative, tumbling target spacecraft in a close-proximity rendezvous scenario. The UKF estimates the relative orbital and attitude states of the target with respect to the servicer based on the pose information extracted from incoming monocular images of the target spacecraft with a Convolutional Neural Network (CNN). In order to enable reliable tracking, the process noise covariance matrix of the UKF is tuned online using adaptive state noise compensation. Specifically, the closed-form process noise model for the relative attitude dynamics is newly derived and implemented. In order to enable a comprehensive analysis of the performance and robustness of the proposed CNN-powered UKF, this paper also introduces the Satellite Hardware-In-the-loop Rendezvous Trajectories (SHIRT) dataset which comprises the labeled imagery of two representative rendezvous trajectories in low Earth orbit. For each trajectory, two sets of images are respectively created from a graphics renderer and a robotic testbed to allow testing the filter's robustness across domain gap. The proposed UKF is evaluated on both domains of the trajectories in SHIRT and is shown to have sub-decimeter-level position and degree-level orientation errors at steady-state.