OPPO was recently a participant at the virtually held Computer Vision and Pattern Recognition Conference (CVPR) 2020. OPPO demonstrated its industry-leading technological strengths and innovative breakthroughs, particularly in the AI space, as well as unveiling a new department of Intelligent Perception and Interaction to explore cutting-edge AI technologies in their Research Institute. OPPO’s success was highlighted by placing in three different contests, winning two first-place and two third-place awards at the CVPR 2020 Challenges.
Technologies that won the awards include: perceptual extreme super-resolution technology which can make a blurry photo 16 times sharper naturally, visual localisation that can tell where the photo is taken by analyzing the image, and human activity recognition in videos
Additionally, members of OPPO’s Research and Development team participated in several sessions of the conference, including delivering keynote speeches and participating in seminars to share their insights on computer vision technologies with thousands of leading researchers worldwide.
CVPR is a premier annual computer vision event organized by the IEEE (The Institute of Electrical and Electronics Engineers). It covers every aspect of computer vision and pattern recognition, machine learning, and AI, and provides exceptional value for students, academics and industry researchers.
Eric Guo, Chief Scientist, Intelligent Perception, OPPO, said “we are very pleased to have achieved remarkable results on behalf of OPPO in this year’s CVPR Challenges, reflecting three of our core research directions: environmental perception, human-centric computing, and image/video generation. This is the first public demonstration of our R&D results since the establishment of the Intelligent Perception and Interaction Research Department of the OPPO Research Institute, which also fully reflect OPPO’s AI strength”.
He added, “we all know that the invention of the mouse and graphical user interface contributed to the success of the personal computer and touch screen deserves much credit for the booming success of smartphones. Every time the human-computer interaction mode and intelligent perception technology evolve, they lead to a brand new tech trend. At OPPO, we are committed to making smart devices more intelligent, so they understand users better. We will then be able to develop a new human-computer interaction method to lead the next wave of technological progress”.
OPPO believes that the cutting-edge exploration in the field of intelligent perception and interaction will fully empower OPPO technologies and products in the future, which will provide users with more innovative interaction method and experience.
OPPO US Research Centre also participated in CVPR 2020 and won the 6th place in Real-World Super-Resolution Challenge, Track 1: Image Processing artefacts. Meng Zibo, manager of Spacetime Lab of OPPO US Research Centre, gave a keynote speech on the institute’s leading research on AI-assisted low light photography titled “Looking into the dark: from image to video” and Shuxue Quan, Senior Director of Computer Vision and AI gave a speech on localisation and mapping titled “Building a unified SLAM system running on both smartphone and AR glass from scratch”
Perceptual Extreme Super-Resolution Challenge
In this challenge, the OPPO Research team was charged with solving the problem of perceptual extreme super-resolution for single image with a magnification factor x 16. The aim is to obtain a model capable to produce high-resolution results with the best perceptual quality and similarity to the ground truth. The team developed and demonstrated the effectiveness of their own neural network called RFB-ERSGAN, beating out 280 other participants with its more intelligent analysis, more efficient computation, and more diversified methods, ultimately winning the first-place among teams from Facebook, Baidu, Microsoft, Amazon, and the Chinese Academy of Science.
OPPO’s RFB-ERSGAN neural network produces more natural textures, e.g., animal fur, building structure and plant texture.
The technology will be applied in a number of scenarios. For example, it can help users to change LR photos taken by their smartphones to HR pictures with rich details. Other applications may include: restoring old or damaged photos and old films; refining important medical images or conducting detailed analyses on satellite images and space images.
Visual Localisation for Handheld Devices Challenge
OPPO develops AI applications to analyse pictures from multiple dimensions
Achieving one first-place on outdoor visual localisation (Aachen Day-Night) and one third-place on indoor visual localisation(InLoc), OPPO created a monocular visual localisation pipeline that leverages semantic and depth cues to find the precise location of any query image. Overcoming a major obstacle for many computer vision applications, which is to identify and eliminate a significant amount of incorrect results, this method has performed at a state-of-the-art level through robust experimentation.
Likely future uses include augmented reality, intelligent robotics, or precise navigation. For example, when the AI assistant is equipped with a camera, it can leverage visual localisation technology to analyse the pictures taken and achieve accurate positioning and thus move to the right location.
Activity Detection in Extended Videos
Winning third place in Activity Detection in Extended Videos, to respond to this task encouraging the design of automatic activity detection algorithms, OPPO built theirs to efficiently identify and reorganise the characters in each frame and recognise dozens of human activities in complex dynamic videos. This technology will play an important role in the fields of human-computer interaction, and sports video analysis in the future. For instance, the future smartphone will be able to automatically identify moving subjects and adjust the camera focus accordingly.
OPPO develops algorithms to efficiently identify and reorganise human behaviour in complex dynamic videos
New Intelligent Perception and Interaction Department
In addition to their performance in the challenges at CVPR 2020, OPPO Research Institute announced it had established a new Intelligent Perception and Interaction Department earlier this year, marking further exploration by OPPO into cutting-edge AI technologies. Its explicit aim is to vigorously promote research and development in the two core areas of intelligent perception and interaction. The department is bringing talent from many top tech firms and research institutes to support their ongoing efforts. It has also invited leading AI expert Eric Guo to serve as the Chief Intelligent Perception Scientist, who has hosted a workshop about drone detection and tracking materials at the CVPR conference.
Eric Guo used to work for Microsoft as a researcher and has conducted multiple industry-leading researches over the years. His computer vision models are widely used in many industry solutions such as image search, knowledge graph, smart cloud, and smart mobility