Report Title: RobustVisIntel: A Road wards Robustness of Visual Intelligence

Report time: 14:00, December 9th

Tencent Meeting Number: 765-384-260

Report Summary:

With the construction of large-scale datasets, the maturity of high-performance computing devices, and the development of deep learning technology, many visual intelligence tasks have made tremendous progress. However, currently most intelligent models are trained on relatively ideal and clean data. These models ignore the influence of various interference factors that exist in the real world, such as brightness imbalance caused by changes in scene lighting, blurring caused by camera imaging parameters, and structural distortion caused by data post-processing. How to evaluate the impact of these interference factors on intelligent models and how to improve the robustness of intelligent models against these interferences is the key to achieving the practicality of visual intelligence models. The report will take video target tracking tasks and motion blur as examples to introduce the team's recent series of work in the robustness evaluation and improvement of visual intelligence, and further introduce the team's main ideas in building a new generation of trustworthy artificial intelligence.

Reported by:

Dr. Guo Qing obtained his doctoral degree from Tianjin University in 2019 and joined Nanyang Technological University in Singapore the same year. Since September 2020, I have been employed as a postdoctoral fellow by the President of Nanyang Technological University in Singapore (with 5 candidates from the Global 500). Dr. Guo Qing's main research direction is computer vision and artificial intelligence security, including video target tracking, adversarial sample attacks and defense, image restoration, etc. It has been published in top conferences and journals, such as IEEE TIP, IEEE TIFS, ACMTOSEM, NeuroIPS, CVPR, ICCV, ECCV, ACM-MM, AAAI, IJCAI, etc. have published over 20 papers and corresponding papers, and won the ICME Platinum Best Paper Award in 2018. Dr. Guo Qing is currently one of the top publications and conferences, including IEEE TPAMI, PC members or reviewers for IEEE TIP, IEEE TIFS, ICLR, NeuroIPS, CVPR, ICCV, ECCV, etc.