3D Pose Transfer with Correspondence Learning and Mesh Refinement

Some previous works require key point annotations to build reliable correspondence between the source and target meshes, while other methods do not consider any shape correspondence between sources and targets, which leads to limited generation quality.