On October 27th, 2019, the 2019 International Conference on Computer Vision (ICCV) was held in Seoul, Korea. Student and teacher representatives from the School of Cyberspace Security of Zhejiang University were invited to attend the conference and present Zhejiang University’s latest advancements in the field of artificial intelligence security. ICCV is a top academic conference in the field of computer vision sponsored by IEEE. This year, it attracted more than 7500 computer vision scholars from 59 countries around the world, including top researchers from renowned universities and well-known companies like Google, Facebook, Huawei and SenseTime. The main topics of this year's ICCV were deep learning, 3D target detection and image video.
Point-Cloud Saliency Maps co-authored by University at Buffalo, the State University of New York and Zhejiang University was accepted by the conference and presented in the Oral session on the afternoon of October 27th, 2019 (acceptance rate was only 4.3%). A Ph.D. student (one of the co-authors) from the School of Cyberspace Security of Zhejiang University introduced the project to the scholars and answered questions on site. The project was guided by Professor Ren Kui of Zhejiang University and featured creative methods to identify critical 3D points in 3D point-cloud data inputs. Based on these points that play a crucial role in the model recognition process, the project proposed a new and effective 3D point-cloud model to combat cyber-attacks, which could serve as a powerful reference for the future application of artificial intelligence in 3D models.
“Person-in-WiFi: Fine-grained Person Perception using WiFi” co-authored by Zhejiang University, Xi'an Jiaotong University, Carnegie Mellon University and the Alibaba-Zhejiang University Joint Research Institute was received as a Poster by ICCV (acceptance rate was 25%). This project proposed a solution for fine-grained person perception based on off-the-shelf WiFi antennas and standard IEEE 802.11n WiFi signals. Compared to existing image-based solutions, this WiFi-based solution is cheaper, unaffected by environmental factors (such as lighting) and has no privacy issues.
With the maturity of deep learning in computer vision applications, technologies such as autonomous driving and 3D detection have developed rapidly, and many large companies have also conducted in-depth research. However, while these applications display extraordinary potential, the key to their wide and secure application in practical scenarios is how we interpret the principles of deep learning to obtain reliable and robust models. The School of Cyber Science and Technology has made innovative explorations in this emerging field and will continue to contribute to the development of artificial intelligence technologies in the future.