Supplementary Material

Abstract

We provide additional material to supplement our work. Appendix A verifies the accuracy of our re-implementation of [1], which is, to our knowledge, the only pre-existing work on few-shot 3D reconstruction. Appendix B further examines ways of incorporating shape priors into the encoder-decoder architecture of [1]. In Appendix C, we report performance on base classes for our three considered methods and Wallace et al. [1]. Appendix D shows learned attention maps obtained using the CGCE model, analyses similarities across classes, and the choice of hyperparameters. Finally, we provide more qualitative examples in Appendix E.