Every image is encoded using an ImageNet pre-trained ResNet18 to a latent feature map z R4 4 256. A flattened version of z is processed with one linear layer with output channels equal to N 3to get the predictions for points u and visibility v. We apply the sigmoid function to the visibility predictions v to enforce a numerical range [0,1]. Our models are trained using Adam optimizer with learning rate equal to 1e-4. In detail, scale is sampled from the range [0.7, 1.2], vertical translation is up to 38 pixels and we also apply 2D rotation up to 40 degrees. For camera equivariance the image is simply flipped horizontally and given as input to the network to estimate the pose.

Information borrowing from historical data is gaining attention in clinical trials of rare and pediatric diseases, where statistical power may be insufficient for confirmation of efficacy if the sample size is small. Although Bayesian information borrowing methods are well established, test-then-pool and equivalence-based test-then-pool methods have recently been proposed as frequentist methods to determine whether historical data should be used for statistical hypothesis testing. Depending on the results of the hypothesis testing, historical data may not be usable. This paper proposes a dynamic borrowing method for historical information based on the similarity between current and historical data. In our proposed method of dynamic information borrowing, as in Bayesian dynamic borrowing, the amount of borrowing ranges from 0% to 100%. We propose two methods using the density function of the t-distribution and a logistic function as a similarity measure. We evaluate the performance of the proposed methods through Monte Carlo simulations. We demonstrate the usefulness of borrowing information by reanalyzing actual clinical trial data.