On February 27, the 2nd China Network Performance (Live) Industry Summit Forum and Webcast Annual Festival was held in Beijing, which is a benchmark for the live broadcast industry. Kingsoft Cloud was invited by its rich experience and industry influence in the field of AI content services. Dr. Su Chi, Chief Algorithm Architect of King Insight AI, delivered a keynote speech at the conference.
"AI Live Content Service" is an important part of Kingsoft's "King Insight" product service. As early as the live broadcast industry is in the ascendant, it has entered the stage of technology research and development, and later came online when the "Internet Live Service Management Regulations" was released. After years of hard work in the market, "King Insight" has become a leader in the field of short video and live content security services. The live broadcast industry is developing rapidly, and the "King Insight" AI content service is also innovating, realizing the "role" transformation from network purifying to content mining, from security guardians to content guides.
Mature industry and advanced AI
The short video and live broadcast industry experienced supervised sharp warnings in "savage growth" and also experienced many fierce competitions. At the moment, the industry has entered a healthy and stable state. As the industry has been more standardized, the Matthew effect has also emerged, and the industry landscape has become clearer. The pursuit of better quality content has become the common goal of all platforms. "Currently, King Insight provides six types of comprehensive audit services for content management, including image review, semantic review, intelligent OCR, intelligent voice, video understanding, and intelligent label. With the trend of diversified layout, overseas expansion, intensive cultivation, short video and live broadcast combination, King Insight has undergone tremendous changes in service types, service models and AI algorithm models. Every small step forward in the industry is a big step in content service products, and it is also a technological breakthrough in AI technology." Dr. Su said at the scene.
With the constant changes in the industry's ecology, structure and development, Kingsoft Cloud adheres to underlying principles and promotes a new industry form to realize the transformation of content industry value. Kingsoft Cloud officially launched the "show live understanding solution", "game live understanding solution" and "short video content understanding solution". Relying on AI content service capabilities, Kingsoft Cloud helps the platform to make refined operation in the entire process, from production to distribution. Kingsoft Cloud has solidified the technical foundation for high-quality content output, breaking the barrier between authors and users, and achieving differentiated layout.
How to make AI have eyes that can appreciate beauty
"Content understanding", as its name suggests, is to let AI have the ability to understand the deep meaning and logical relationship of people and things. Relying on the patented time series algorithm of Kingsoft Cloud, the live video and short video content understanding solution realizes high-accuracy recognition by constructing a dynamic video processing model, including video category, host style, scene and object recognition.
Take the show host as an example. In addition to identifying his or her appearance and personal identity, Kingsoft Cloud's AI content understanding service also needs to understand his or her personal style, such as loli, cute, handsome or mature, and his or her current behavior and environment. Through fine label classification, it can provide services for differentiated and refined operation of live broadcasting platform and tapping potential value content.
The Kingsoft cloud content understanding model currently supports more than 100 video categories, such as walk the dog, dance, beauty, food, fitness, music, games, sports, and landscapes. The model supports fine-grained tags that are accurate to the game name, dance type, song style, and more. Giving AI the ability to "find a better life" is the result of the research and development of King Insight AI algorithm team, which is, training thousands of high-quality short videos, and obtaining a content understanding model with strong generalization ability, as well as the patented time series algorithm.
According to Dr. Su, content understanding is not a general image recognition, it must fully capture the time series information in the video content. The Kingsoft Cloud content understanding model uses 3D conv and LSTM and Attention to finely capture the partial and whole information of video single-frame images. Inspired by the brain's attention mechanism, the AI algorithm team introduces a "time-space attention mechanism", which enables the model to focus on key frames and key position information, and reduce the impact of unrelated frames on model performance. The entire model does not require any manual intervention. By inputting the original video, you can get the final predicted result with high precision and speed. At present, King Insight content understanding model only takes 30 milliseconds to process a single video with an accuracy of more than 80%.
With the advent of the 5G era, lasting industry competition will stimulate content platform participants to continuously carry out technological innovation. Relying on more refined and powerful operations, users and high-quality content producers are closely linked. This will continue to promote the prosperity of the cultural industry. As a participant, King Insight's one-stop AI content service products will have a higher market value.