Analysis of Failure Cases in Our Image Generation Method

Our method exhibits three primary types of failure cases. The first type, as depicted in Fig.//ref;fig1//(a), occurs when the target character's hands are clenched, thus resulting in inconsistent hand poses between the generated source character and the target character. The second type arises due to the presence of artifacts, such as arms and calves, in the input image. Consequently, this leads to the generation of source images with missing forearms and misaligned calves, as illustrated in Fig.//ref;fig1//(b). The last type of failure occurs when a third person appears in the target person video, causing the failure to generate the source person image. In such cases, the system mistakenly assumes the necessity of learning the action of transforming the third person, as shown in Fig.//ref;fig1//(c). Therefore, it is crucial to address these failure cases in order to improve the performance of our method.