In recent years, novel view synthesis has performed well in various research field. Previous works solve the problem by using additional 3D information, but that are difficult to apply in real scenarios. Thus, we mainly research on how to generate novel view for a single image via key points.

However, the pose variation of vehicle images is one of the key challenges. Same vehicle identities with different viewpoint usually have large discrepancy. In this paper, we propose a method based on Generative Adversarial Networks (GANs) to generate fake images that have the same viewpoint to solve the different pose problem. Our proposed PTGAN first extracts identity-related and pose-unrelated representations from input images and then concatenates the representation with the pose information to generate the fake image with the assigned pose to deal with the pose variation problem.

Github: https://github.com/fan84sunny/PTGAN