MetaSDF-Supplementary Material-Eric R. Chan

These authors contributed equally to this work. Quantitative comparison for models trained on ShapeNet V2 Tables and evaluated on ShapeNet V2 Benches..................... Here, we demonstrate that conditioning via concatenation is a special case of a hypernetwork, where the hypernetwork is a single affine layer that only predicts the biases of the hyponetwork. We first formalize a hypernetwork Φ that predicts the weights of a single layer of some hyponetwork. We are only interested in the weights and biases and therefore omit the nonlinearity. This can be formalized as follows: y = W (x z) + b (2) where () signifies concatenation.