Newsroom

Every new move is a step forward.

Newsroom

SenseTime Advances Global Computer Vision Development with 62 Papers Accepted to CVPR 2020

2020-05-12 10:09

                                                                                                                 

Beijing – May 12, 2020 – SenseTime, the world-leading artificial intelligence (AI) company, is proud to announce that 62 papers authored or co-authored by SenseTime were accepted to the Conference on Computer Vision and Pattern Recognition (CVPR), among the world's most important computer vision conferences. This further reinforces SenseTime’s goal to advance global computer vision development after its algorithms also topped the recent National Institute of Standards and Technology’s (NIST) Face Recognition Vendor Test (FRVT) earlier in March.

SenseTime’s papers explore and address some of the most critical industry challenges in computer vision today, including Generative Adversarial Networks (GANs); 3D object detection through 3D Point Clouds understanding and detection; accelerated data training and quantized detection models; image/video understanding and analysis (IVUA) and Neural Architecture Search (NAS). These research papers make a significant contribution to the ongoing development of AI industry.

One ground-breaking development is the study of GANs which looks into the latent space of GANs for semantic face editing. Based on a simple and general technique called InterFaceGAN, SenseTime’s technology manages to control the pose as well as other facial attributes, such as gender, age, eyeglasses. More importantly, SenseTime is able to correct the artifacts made by GANs. This approach analyses the latent space of GANs to better convert random noise into a high-quality image.  

Accelerating the training process of deep learning has always been another major challenge in the AI industry. One of SenseTime’s accepted papers at CVPR 2020 proposes the use of an 8-bit numerical training model to address the technologies for INT8 training, speeding up the training process for convolutional neural networks. The proposed method also manages to reduce calculation loss and achieve lossless training accuracy. In addition to being accepted to the CVPR, SenseTime’s algorithms were ranked number one for the highest accuracy at five of the NIST’s FRVT 1: N categories in March this year.   

                                                          

The results from CVPR 2020 once again demonstrated SenseTime’s research capabilities in the field of computer vision. As a world-leading AI company, SenseTime will continue its research mission to innovate and develop AI technologies to empower various industries and businesses for a better tomorrow.

The CVPR 2020, co-sponsored by the IEEE Computer Society and the Computer Vision Foundation, received a total of 6,656 papers this year, among which only 1,470 papers were accepted, one of the lowest acceptance rates in ten years. Despite demanding selection criteria, SenseTime successfully admitted 62 papers to the conference.