On the afternoon of September 8, the first "Malan Mountain Cup" International Audio and Video Algorithm Optimization Competition Awards Ceremony and Summit Forum was held in Malan Mountain. This event is an important part of the 2020 Internet Yuelu Summit.
Cutting-edge topics originate from industry needs
Malanshan Video Cultural and Creative Industrial Park has gathered a large number of audio and video related enterprises. In the common people's impression, the audio and video industry belongs to the category of cultural creativity, but in fact its development cannot be separated from the support of science and technology. In May of this year, the park and Mango TV jointly held the first "Malanshan Cup" International Audio and Video Algorithm Optimization Competition, which challenged 3 cutting-edge topics in the audio and video field, including video specific point tracking, video recommendation, and video quality Damage repair.
These three topics are all based on the needs of industry development. Take video tracking at specific points as an example. What it wants to solve is the problem of ad placement. This is a problem that has plagued video sites, advertisers and viewers for a long time.
Traditionally, video sites will insert a post-advertising before the feature film begins. After the audience chooses to spend money to become a member, they can skip this advertisement and watch the feature film directly. But this business model has a flaw. It puts advertisers and viewers on opposite sides. The higher the proportion of viewers becoming members, the worse the advertising effect. The result is a structural contradiction between advertising revenue and membership revenue.
In order to solve this contradiction, video websites choose to replace the patch ads with embedded ads. The embedded ads themselves are an organic part of the feature film, such as the drinks the guests drink in variety shows, the mobile phones used by the characters in TV dramas, etc., can attract relevant Advertisers. However, this business model relies on the investment in the early stage of the project, which poses a certain risk to advertisers, just like a bet. There may be TV shows that advertisers love, but audiences don't buy it, or advertisers don't like variety shows, but audiences are very popular. This is more obvious for new programs that are particularly innovative. Sometimes a good idea can only be kept deep in my mind because there is no "golden father" to pay the bill.
Is there any way to save viewers from the suffering of SMD ads and reduce the risk of decision-making by advertisers? In February last year, a video of Yang Mi changing faces and Zhu Yin became popular on the Internet, providing a solution-since the dynamic faces in the video can be replaced, can the drinks be changed? Can the phone be changed? Even if there were no advertisements in the early stage of investment promotion, after the broadcast, the response was good, and the advertisers would come to the door and replace them with the corresponding brands. To do this, it is not enough to cut out the image frame by frame. The workload is too much, and you can only rely on the power of technology to let the computer replace it automatically. This is the industry demand that the video specific point tracking must meet.
Broad application prospects for research results
After the announcement of the "Malanshan Cup", it immediately attracted the attention of a large number of talents from universities, research institutes, and Internet companies. A total of 1,294 teams signed up for the competition, including 34 from Peking University, 25 from Tsinghua University, and 37 from top foreign universities such as Massachusetts Institute of Technology.
Dr. Zeng Hui from Hong Kong Polytechnic University and his friends formed the ManGoGo team to jointly challenge the video quality damage repair project. Zeng Hui's supervisor has been conducting research in the field of image quality for more than ten years and has published many papers. In Zeng Hui's view, the purpose of scientific research is not just to publish papers, he also hopes that what he studies can help solve practical problems. This is what attracts him most about the "Malanshan Cup". He can walk out of the laboratory and understand the real needs of the industry. This is an important reason why he signed up for the competition. "The image quality compression and noise in the topic of this competition are frequently seen in the industry, which provides us with many real-world experience and data support for our future research."
In Zeng Hui's view, the application prospects for video quality damage repair are very broad. "The country attaches great importance to the development of 4K ultra-high-definition television. The State Administration of State Council issued the "4K Ultra-High Definition Television Program Production Technology Implementation Guide (2020 Edition)" in June. In the context of policy advantages and popularization of technology, relevant application scenarios and demands will greatly increase."
In addition to looking forward, repairing video quality damage can also look backward to restore old image data. Zeng Hui gave an example: "Like some classic old dramas, there will be problems with 4K ultra-high-definition TV broadcasts, so they need to be repaired." Classics such as "Returning the Pearl" and "Soldier Assault" have all gone through such repairs before they are restored. Meet the audience. In addition, repairing old images is also of great significance for preserving historical memory and inheriting traditional culture. For example, in the 1980s, Hunan recorded performances of local operas such as Xiang Opera and Huagu Opera. However, due to the technical conditions at the time, these videos had low pixels and high noise. The facial expressions of the actors could not be clearly seen and they could not be heard clearly. The details of the actor's singing and reading can only be vaguely seen, which is the regret in the hearts of many fans. After the restoration, the clarity of these opera data will rise to a higher level, which is of great significance to the inheritance of opera.
Because of this, Zeng Hui said with certainty: “It can be said that video quality damage repair has rich practical scenarios for both the future and the past.”
Attract algorithm talents to join Changsha
In addition to meeting the actual needs of the industry, holding the "Malanshan Cup" has another important purpose-to attract a group of outstanding algorithm talents to Changsha and join the Internet industry. At the forum, Wang Xinying, Luo Jiangwei, and Wang Wei signed contracts with Mango TV. They formally became "new Changsha people" under the banner of similar interests.
When the reporter saw Wang Xinying, the Xiang sister Yao Ting immediately came to mind. The latter was selected for Huawei's "Genius Boys" project in August and attracted a lot of likes on the Internet. Wang Xinying is also such a highly educated beauty. After graduating from University College London, she came to a company in Fuzhou to work as an algorithm engineer. It was an accident that she became involved with Ma Lanshan.
Wang Xinying took a course in medical image registration while in London. Now there are many testing equipment in hospitals, such as X-ray, B-ultrasound, CT, NMR, etc. Each device can produce a picture. Is there a way to combine the pictures made by these different equipment into a general picture, which is better To judge the condition? Medical image registration is to solve this problem.
After seeing the topic of "Malanshan Cup", Wang Xinying keenly noticed that the problem of video tracking at specific points can be solved by medical image registration. The essence is to use the principle of spatial geometric transformation in mathematics to convert the first frame of the video. The target is mapped to each subsequent frame. So she formed the fried rice cake team alone to participate in the competition and won the championship with an absolute advantage of MSE0.588, which means that with her algorithm, the average coordinate point pixel error can be controlled within 1 pixel.
Wang Xinying's method of transferring expertise in different fields made the contest organizers shine. She told reporters: "I am not necessarily better than other players. The key to breaking through may be the courage to break the inherent thinking mode, think and apply flexibly, and try to combine across fields."
Wang Xinying is such a girl who has the courage to try new things. Changsha is a completely unfamiliar city to her, and the video industry is a field that she has little contact with. With the sincere invitation of the organizer, she finally decided to leave Fuzhou and devote herself to the embrace of the Internet industry in Changsha. Talking about why she came to Changsha, she said: "Interest is the best motivation. I participated because I was interested in this project and researched technology. The combination of artificial intelligence and audio and video is an endless sea of stars. Changsha gave me this. The opportunity allows me to ride the wind and waves in the ocean of algorithms of interest."
Luo Jiangwei also came to Changsha because of interest. He was not from a subject, and he proved his ability through this competition and got the coveted algorithm position. He encouraged more young people: "Don't give up lightly, stick to what you love, one day we can get together in Malan Mountain."
(Source: Changsha Evening News)