Neural representations are popular for representing shapes as they can be used for data
cleanup, model completion, shape editing, and shape synthesis. Current neural
representations can be categorized as either overfitting to a single object instance, or
representing a collection of objects. However, neither allows accurate editing of neural
scene representations: on the one hand, methods that overfit objects achieve highly
accurate reconstructions but do not support editing, as they do not generalize to unseen …

In this supplementary document, we provide additional details on our data preparation
procedure (Section S2), our architecture (Section S3), and the baselines (Section S4).
Additionally, we provide a quantitative evaluation of the shape editing experiments (Section
S5), extend the shape mixing experiments to include a comparison to COALESCE (Section
S6), and provide several additional ablations of our approach (Section S7).