- Every new move is a step forward.
June 22, 2022 – SenseTime is continuing to lead its momentum in computer vision research by bringing technology innovation with a record-breaking 71 papers accepted to this year’s conference on Computer Vision and Pattern Recognition (CVPR), the world’s largest annual conference on computer vision and AI that takes place on 19-23 June, 2022 in New Orleans, Louisiana, USA.
Nearly a quarter of the papers submitted by SenseTime were accepted in oral presentations, covering many frontier areas in both AI research and the industry including 3D vision and autonomous driving.
Prof. Wang Xiaogang, Co-founder, Chief Scientist and President of Intelligent Automotive Group at SenseTime, said, “As researchers, the number of papers and competition rankings were not key performance indicators for us. Our aim is that by providing a favourable environment to our research team, young researchers feel encouraged and empowered to do high-quality research to solve real-world challenges.” Since its establishment, SenseTime and its joint labs have published more than 700 papers in top-tier academic conferences and won more than 70 world championships in various competitions, leading in the computer vision field globally.
In addition to the outstanding number of papers accepted, SenseTime participated in several CVPR challenges including Embodied AI 2022, in which the company successfully won the championship at the RxR-Habitat challenge track together with the Institute of Automation, Chinese Academy of Sciences and Shanghai Artificial Intelligence Laboratory as a joint team. During the contest, the joint team improved the navigation results with natural language control for indoor robots by over 90%. Moreover, in this year’s Challenge on Learned Image Compression, SenseTime’s team won the title of Image Compression Track.
Leading An Industry Breakthrough with Cutting-edge Technologies
Leveraging on its AI infrastructure SenseCore, SenseTime’s research team has been proposing innovative ideas that can be applied in various scenarios to address the real-world challenges.
For example, in the paper Bailando: 3D dance generation via actor-critical GPT with musical memory, researchers put forward a new music-to-dance framework Bailando, which can drive 3D characters to make freestyle dance movements with music, bringing more intelligent and personalized character creation to the digital human industry.
Also in the paper PTTR: Relative 3D point cloud object tracking with transformer, SenseTime’s research team proposed a novel framework for 3D point cloud single object tracking, which significantly improved the accuracy of target tracking on multiple data sets and laid a foundation for the safe operation of autonomous driving.
PTTR: A novel framework for 3D point cloud single object tracking
In addition, SenseTime and Prof. Liu Xianglong from Beihang University together launched the Robust Models towards Open-world Classification, which aims to promote trustworthy AI models. The challenge attracted 286 teams and 416 participants. The winners were announced at the CVPR2022 Art of Robustness Workshop on June 19.
SenseCore AI Infrastructure Lays Solid Foundation for Technological Breakthroughs
SenseTime’s achievements in the research field are accomplished through the relentless innovation and improvement of proprietary technologies which are supported by SenseCore, its universal AI infrastructure that drives the industrialization of AI model innovation and production.
Since 2017, SenseTime has been developing a broad-based ecosystem through joint efforts with the AI industry, including the establishment of Global AI Academic Alliance (GAIAA), which includes members of top universities worldwide to promote global research collaboration and develop research talents.
In addition, SenseTime continues to open-source its technologies to empower the academic and industrial ecosystem. Following the upgrade of OpenMMLab, the AI algorithm open-source platform, SenseTime further open-source OpenGVLab and released OpenDILab and "INTERN", a new generation of general vision technology system together with Shanghai Artificial Intelligence Laboratory and other institutions. As of today, the OpenMMLab has successfully open-source thousands of models to researchers, with more than 50,000 stars obtained on GitHub.