Recently, the annual academic seminar RACV (Recent Advances on Computer Vision) of the Computer Vision Committee of the China Computer Federation (CCF-CV) was successfully held in Dalian. The opening ceremony of the conference was presided over by Professor Yu Jingyi, deputy director of the special committee and professor of ShanghaiTech University. Researcher Tang Weiqing, secretary-general of China Computer Federation, and professor Lu Huchuan of Dalian University of Technology delivered opening speeches, from Tsinghua University, Peking University, Shanghai Jiaotong University, Chinese Academy of Sciences and other universities , researchers from academic institutions, and technical experts from technology companies such as Baidu, Huawei, Microsoft Research Asia, Aerospace Hongtu, and Hehe Information were invited to participate in the discussion.
Group photo of RACV2022 guests
The Computer Vision Special Committee of the China Computer Federation (hereinafter referred to as the "Special Committee") is a professional branch of the computer vision field directly under the China Computer Federation. Its goal is to better develop academic/technical research based on the professional content of the computer vision discipline. Exchange and develop strategic research, promote understanding and cooperation among domestic scholars, promote the development of domestic computer vision disciplines, and enhance the influence of my country's computer vision research in the international field. RACV is positioned as a small-scale high-quality seminar in the field of computer vision in China. It aims to deeply explore potential problems in related research fields through pragmatic, open and equal dialogue and discussion, and provide researchers with new perspectives and perspectives for observing problems. .
According to the preliminary discussion votes of the Standing Committee of the special committee, this meeting set up 3 core discussion themes. In the seminar of theme 1 "Basic Vision Model", the participants explored mask image modeling, self-supervised representation pre-training, and the role of language on the basic model of vision; theme 2 "3D reconstruction and immersive rendering" Focus on the academic frontier and development direction of neural network rendering and generation technology, 3D reconstruction and immersive rendering, business prospects and potential risks in the context of metaverse; theme three is "embodied vision", which mainly discusses the core of embodied intelligence Scientific issues and their role in the field of computer vision, technical routes, evaluation standards, how to promote the implementation of intelligent robots and other issues.
During the theme discussion session, experts and scholars conducted in-depth exchanges on the above topics. Dr. Ding Kai, Director of Natural Language Algorithm Research and Development of Hehe Information, and Dr. Guo Fengjun, Director of Image Algorithm Research and Development, also focused on the two aspects of "visual basic model", "3D reconstruction and immersive rendering". This topic expresses opinions, and proposes that when the visual basic model is integrated with other modalities, we can focus on how to integrate with existing structured knowledge to improve the understanding ability of the visual basic model. In addition, technologies such as 3D reconstruction and immersive rendering are currently more focused on visual reconstruction in the context of the metaverse. How to better combine these technologies with human common sense and knowledge to realize the vision of smooth communication and interaction between people in the virtual world will become a new addition to the future commercialization prospects.
Hehe Information is an artificial intelligence and big data technology enterprise. Since its establishment in 2006, it has accumulated more than ten years of research in the field of intelligent text recognition applications. It is reported that Hehe Information focuses on intelligent document image processing, complex scene document recognition, handwriting erasure and other technologies. In the past three years, it has won the championship in 15 artificial intelligence international competitions such as ICDAR and ICPR, and its academic achievements have been published in CVPR and AAAI. , ACL and other top meetings. Hehe Information uses a number of innovative technologies to significantly improve the performance and effect of text recognition and understanding in complex scenes. Related projects have won the second prize of the China Society of Image and Graphics (CSIG) Science and Technology Progress Award, and have been applied to C-end, B-end products and industry solutions provide digital and intelligent services to hundreds of millions of users in more than 100 countries and regions around the world.