Human Motion Transfer: Clarification and Model Specialization

Thank you for the valuable feedback! Firstly, we would like to clarify that the desired poses are obtained from various source persons, rather than a single source person. Additionally, the given images correspond to different target persons, as shown in figures 6 and 7. Secondly, it is important to note that the source person and the target person are not necessarily the same individual, and they may even differ in terms of gender. Furthermore, as addressed in our response to Q2, our method falls under the category of a specialized-purpose model. Therefore, as illustrated in figures 6 and 7, we have trained a dedicated model for each source person. The figures present a list of 10 source persons, thereby showcasing 10 training models. Figure 6 illustrates a quality comparison among different methods for the same desired pose, while Figure 7 showcases the generation effect achieved by different models (corresponding to different source characters) using the same set of desired poses. Hence, in theory, by employing the existing model for any desired poses originating from any target persons, one can accomplish the transfer of human motion from the target person to the source person through re-inference.