Computer Vision and Pattern Recognition Conference-CVPR 2021
The world’s leading smart product brand participating in OPPO, with its successful work in the field of artificial intelligence, from 7 different competitions returned with 12 awards. These included 1 first place, 7 second place and 4 third place awards, demonstrating the company’s industry-leading technological strengths and innovative breakthroughs in artificial intelligence.
OPPO in the competition, last year, Research Institute represented the Intelligent Sensing and Interaction Department, which was established to explore the latest artificial intelligence technologies. Through the optimization and training of AI algorithms, the work of the teams continues to strengthen OPPO’s AI capabilities and the ability of this technology to better serve people.
Eric Guo, Chief Engineer of OPPO Intelligent Sensing Unit, said the following about the competition: “Last year, we won the first prize in the Perceptual Extreme Super-Resolution category of the competition with our technology that can sharpen the appearance of blurry images. We are very happy that we achieved outstanding results in the competition this year as well. Multi-Agent Behavior, AVA- 3D Face Recognition from Kinetics and 2D Images technologies cover the more complex and advanced areas of computer vision, including behavior detection, localization of human actions in space and time, and face detection, all of which cover many industries such as manufacturing, home, office, photography, healthcare and mobility. It can be used in various scenarios. As OPPO, we aim to enable artificial intelligence to serve people better, and to offer smarter and more useful experiences to users.”
OPPO won 3 awards among 11 awards in the Multi-Agent Behavior category of the competition, which evaluates the ability of an artificial intelligence model to understand, describe and predict complex interactions between animals and humans. It won the first prize in the Learning New Behavior category, the second prize in the Classical Classification, and the third prize in the Annotation Style Transfer.
OPPO improves AI’s ability to understand humans
Developing capabilities in the field of human-oriented artificial intelligence with the mission of ‘Technology for Humanity, Good for the World’, OPPO succeeded in recreating 3D face shapes with an error of about 1 mm with its own artificial intelligence algorithm in 3D Face Identification technology from 2D Images, and ranked second in the main index score ranking. placed in the queue. OPPO’s face recognition algorithm can identify 635 key feature points 30 times per second. In the future, this algorithm will support the development of portrait video technology with 3D feature recognition that makes beautification filters look more realistic and personalized, allowing for richer and more seamless augmented reality filters on social platforms.
Artificial intelligence that understands space and time
OPPO’s AI capabilities have reached the stage where they can recognize human actions in space and time. In the SoccerNet category of the competition, OPPO took second place in both the Action Spotting and Replay Grounding missions. The purpose of the task was to evaluate the algorithms’ ability to identify more than a dozen actions in a football match video, such as offside and red card violations, which are difficult to understand due to the complexity of the rules. In the future, these AI capabilities can also be used to analyze technical matches for football fans and automatically generate highlights from a user’s life, such as weekly highlight clips, by analyzing videos on their smartphones.
Ranking second in both the Cross-Modal Action Recognition and Cross-Model Action Temporal Localization tasks in the MMact Challenge, OPPO’s powerful AI algorithm can accurately track more than ten action types in a video, such as talking, crouching and walking, using only visual data. can recognize. This technology could provide many benefits in the future, including the ability to better care for children, pets or the elderly in smart homes. For example, an infant or child can alert parents in another room the moment they are performing potentially dangerous actions.
OPPO also placed third in the AVA-Kinetics Challenge, which uses the industry’s first dataset containing both space and time information. The AVA-Kinetics algorithm not only accurately determines the various behaviors of the people in the video, but also notes their time and location. As a result, OPPO’s artificial intelligence technology understands not only what you do, but also where and when you do it.
OPPO continues to explore the limits of artificial intelligence technology
At this year’s CVPR, OPPO has achieved new success in further academic challenges by winning 2 third prizes in the competition’s LOVEU (Long-form Video Understanding) category. The LOVEU contest requires artificial intelligence technology to understand the content. Considering the variety of content, this category of the competition is of great importance for the application of artificial intelligence algorithms to more generalized situations. In the future, this technology will have the potential to be widely used in other AI tasks such as face detection and behavior recognition in video processing.
Demonstrating its technology to produce 3D depth based on 2D images, OPPO USA Research Center took the second place in the Dense Depth for Autonomous Driving Competition and took the Innovation Award home. Using deep learning models to extract depth information from original images, this technology could replace depth sensors such as ToF to deliver better indoor and outdoor navigation experiences in the future.