Organised by the CPC Hunan Provincial Committee Cyberspace Administration, Hunan Association for Science and Technology, China (Changsha) Malan Mountain Video Cultural and Creative Industrial Park, and Mango TV, the first "Malan Mountain Cup" International Audio & Video Algorithm Competition was successfully concluded on 15 June 2020. It attracted nearly 1300 teams from over 20 countries and regions including China, USA, UK and Australia, to compete on the AI algorithm technologies in 30 days.
We are delighted to announce that “ManGoGoGo”, a team consisted of our PhD students Mr Hui Zeng and Mr Xi Yang, together with Mr Hui Xian from DAMO Academy of Alibaba Group, won the Champion under the track of “Image Quality Restoration” in the competition. The team proudly received the honour at the award presentation ceremony held in Changsha, China on 8 September 2020.
“We joined the contest because we are very curious about the needs of the industry. Our goal is not just to have the research papers published but also hope that our research work can solve real-world problems,” said Mr Zeng, the team leader. “We have laid a solid foundation through our research studies supervised by Prof. Lei Zhang. The training enabled us to identify a clear direction upon receiving the competition data, allowing us to save much time and effort in completing the tasks.”
“ManGoGoGo” solved the challenge of restoring a damaged video to a high definition one by adopting a multi-frame strategy to fully utilize the complementary information among video frames, which could significantly improve the signal to noise ratio. The team designed a network architecture taking large image patches as inputs, facilitating the model to leverage the larger scale of spatial information for a better restoration. They also carefully controlled the model complexity to balance its performance and efficiency.
“The prospects in the image quality restoration field are quite broad. Under the national policy of advancing and popularising the 4K ultra high definition technology, the application and demand for image quality restoration are expected to increase greatly. For example, some TV dramas and advertising video products are requiring higher image quality,” another team member Mr Yang shared. “On the other hand, the restoration technique is often used to repair some old data and images, which is also noteworthy for cultural heritage.”
The “Malan Mountain Cup” comprises three tracks: Video Specific Point Tracking, Video Recommendations and Image Quality Restoration. The competition aims at building a technology platform for top algorithm talents to demonstrate their innovative ideas and skills of deep learning as well as to explore the future of the film and television industry.
Some contents above are extracted and translated from the interview by Toutiao of Mango TV
Photo source: Mango TV
