With the development of neural networks, high-quality images can be generated automatically by different algorithms with semantic information. Go further, it is also possible to generate a video with few sentences or other semantic information.
In this project, when designing the generative model (e.g. Generative Adversarial Network), it is not only necessary to consider the location/shape/texture of the objects in each frame, but also to utilize the temporal dependences and constraints.
It is a challenging but potential work. Please send me email if you have interest and questions.
Key words: Generative adversarial network (GAN), video generation/video synthesis, semantic analysis, temporal dependencies, recurrent network.
Contact person: Yuren Cong