Description of field of research:

Content creation is a growing field in artificial intelligence (AI). Content could refer to a picture, audio, text, or video. One main motive for research in content creation is how it can benefit a diverse range of sectors and fields by increasing their efficiency. Content creation achieves promising results using a generative model called generative adversarial network (GAN). While generating images using GAN has attained advanced applications, video generation is still at an early stage. One of the main challenges of video GANs is lacking a sufficient amount of data samples to train a model. Data augmentation is a possible approach that overcomes the limitation in the dataset size. In this project, we augment the video GAN pipeline with image samples to create videos that are more diverse and have better spatial resolutions. Expanding a video dataset with image samples might overcome the need for more video samples in datasets where there are obstacles when collecting video samples such as the availability of the samples.

School

Computer Science and Engineering

Research areas

Affective Computing, Virtual Reality, Artificial Intelligence

This project is in the generative models' domain. The successful candidate will work under the supervision of Dr. Gelareh Mohammadi and HDR student Nuha Aldausari from the School of Computer Science and Engineering. This is an opportunity to work through a deep learning lifecycle, from creating a dataset to training a model.

This project involves collecting an image dataset for an existing video dataset. Then, a generative model can be developed to analyze the effectiveness of expanding a video dataset with images. Publishing high-quality outcomes can be possible while optional.

  • https://arxiv.org/pdf/2008.01999.pdf
  • https://www.springerprofessional.de/en/phonicsgan-synthesizing-graphical-videos-from-phonics-songs/19652484
  • https://openaccess.thecvf.com/content/ICCV2021/papers/Gu_LoFGAN_Fusing_Local_Representations_for_Few-Shot_Image_Generation_ICCV_2021_paper.pdf