基于扩散模型的大规模生成式视频生成模型

As OpenAI introduces Sora, a generative text-to-video diffusion model,

it opens the door to generation of high-definition videos at the minute

level from text descriptions. This groundbreaking model not only showcases

the vast potential of video generation but also captures the attention

of researchers and enthusiasts alike. In this talk, we will delve into

the evolutionary path of large-scale video generation models and explore

key research milestones. We will then analyze the technical advancements

and breakthrough effects achieved by the Sora model. However, while the

success of Sora, there still exist limitations and bottlenecks in video

generation. Concluding the talk, we will discuss the challenges faced

in video generation and explore potential avenues for future breakthroughs.

Dr. Xinyuan Chen is currently a researcher at the Shanghai Artificial

Intelligence Lab, collaborating closely with Prof. Yu Qiao. During 2020-2022,

she did her post-doc research at East China Normal University, supervised

by Prof. Yue Lu. In 2020, she completed her dual PhD from Shanghai Jiao

Tong University and the University of Technology Sydney, under the supervision

of Prof. Xiaokang Yang and Prof. Dacheng Tao. Her research interests lie

in generative models, diffusion models, and generative adversarial networks.

Currently, she focuses her work on image and video generation, large-scale

video generation models, as well as controllable generation incorporating

multi-modality and semantic conditions.