inference

[yaping1119]

NeuralPull can learned multiple shapes. use a onehot vector to represent which model to reconstruct.
Thank you for your excellent work. Here's a question I want to consult you.

1. On generalization. Based on NeuralPull's work, PCP can learned multiple shapes. onehot vector is used to represent the model to be reconstructed. In the training local learning phase, multiple shapes can be learned together. In the testing global phase, a prediction query can be learned for each individual point cloud. Can this work be generalized to shapes that have not been seen? Can it be generalized in the local phase, and can it be trained again in the global phase, or does it not support generalization.
2. Questions about datasets. Based on the first question, PCP and NeuralPull datasets, such as ABC data sets, do you think that the quantitative comparison given in your paper was completed on 100 testsets of abc (abc/abc_noisefree/abc_extra_noisy).
   Best wishes, thank you!