Hong Kong – August 26, 2020 – SenseTime, the world-leading artificial intelligence (AI) company, is proud to announce that 60 papers authored or co-authored by SenseTime were accepted to the biennial European Conference on Computer Vision (ECCV) that is currently running virtually through 28 August 2020. This marks another significant milestone in SenseTime’s journey to discover new possibilities for computer vision and advance global development after achieving remarkable results at CVPR 2020 earlier this year.
SenseTime’s papers explore and address some of the most critical challenges in computer vision today, including Generative Adversarial Networks (GANs); 3D Point Cloud Understanding and Analysis; Video Understanding and Analysis, and Object Detection. The research papers once again demonstrate SenseTime’s continuous effort to break new ground through research and innovation as well as contribution to the ongoing development in the field of computer vision.
Among the accepted papers, one of the highlights was an oral presentation about a study exploring the capabilities of a GAN to restore missing semantics such as color, patch, resolution of various degraded images. The GAN also enables diverse image manipulation, including random jittering, image morphing and category transfer.
Another paper of SenseTime focused on improving scene text recognition performance using the RobustScanner to dynamically enhance positional clues for robust text recognition. This approach can effectively avoid garbled text which is likely to occur when using SAR model to perform scene text recognition on contextless application scenarios.
In addition to the outstanding paper acceptance rate, SenseTime also topped the Large Vocabulary Instance Segmentation (LVIS) Challenge which was introduced by Facebook AI Research (FAIR) back in 2018, and the Advances in Image Manipulation workshop (AIM 2020) at ECCV 2020.
SenseTime applied a two-stage fine-tuning approach to improve its algorithms and overcome class imbalance problems and achieve higher quality mask prediction, which in turn promotes efficiency and accuracy of long-tail object recognition. As a result, with its proprietary platform SenseSpring, SenseTime was able to deliver a flexible, efficient and accurate training model which nailed the LVIS Challenge.
At AIM 2020, the XPixel team co-formed by SenseTime and Shenzhen Institutes of Advanced Technology (SIAT) was able to improve motion modelling in video generation and optimize the video temporal super-resolution technology. The team delivered a model that aligns with the actual motion trajectory, further enhancing the smoothness of videos.
As one of the world’s top-tier conferences in computer vision, ECCV received a total of 5,025 paper this year, double the number of the previous conference in 2018, among which only 1,361 papers were accepted. The results from ECCV 2020 demonstrated SenseTime’s research capabilities in the field of computer vision. As a world-leading AI company, SenseTime will continue its research effort to innovate and develop AI technologies to empower various industries and businesses for a better tomorrow.