Abstract 3D pose transfer that aims to transfer the desired pose to a target mesh is one of the
most challenging 3D generation tasks. Previous attempts rely on well-defined parametric
human models or skeletal joints as driving pose sources. However to obtain those clean
pose sources cumbersome but necessary pre-processing pipelines are inevitable hindering
implementations of the real-time applications. This work is driven by the intuition that the
robustness of the model can be enhanced by introducing adversarial samples into the�…

Multi-scale Masking Feature Extractor. The architecture of the feature extractor is presented
in Table 1. The feature extractor is used to extract a set of latent embeddings from S-scale
representations from the given source pose for further mesh generation with the following
decoders. Basically, it works as a normal 3D feature extractor that encodes the input meshes
with vertex number N1 into a latent vector with C size. The difference is that the latent vector
will be expanded up based on the target meshes along the topology dimension (vertex�…