On the evenings of October 23 and 25, the School of Computer Science (SCS)'s 1st @ World "Touching Academic Frontiers" international exchange event of the 2023 autumn semester was successfully held. The @World "Touching Academic Frontiers" international exchange series activities aimed to promote academic exchange, improve academic expression ability, broaden academic horizons, stimulate academic innovative thinking, and build an open and diverse academic exchange and discussion platform.
This event invited three outstanding students from the SCS, Guo Pengcheng, Wu Qiao, and Ren Chunlin, as guests to share their research results.
Guo Pengcheng gave a report titled "ICASSP23 Participation Experience and MISP Challenge Participation System Description". In ICASSP23 (International Conference on Acoustics, Speech and Signal Processing), Guo's team won second place in the MISP Challenge (Multimodal Signal Processing) speech recognition track. Guo Pengcheng shared the basic principles and algorithms of multi-microphone audio processing, how to process large amounts of audio datasets, and the process of his continuously improving and optimizing solutions for specific competition problems, which has provided many inspirations to the students participating in the salon.
Wu Qiao shared his research progress in 3D single target tracking and his experience in attending the International Conference on Computer Vision (ICCV23). 3D single target tracking is a computer vision technique that involves spotting a target in a series of images (or point clouds) and then predicting its position in subsequent frames. In practical applications, such as autonomous driving, 3D single target tracking can help the system understand the environment more accurately. Compared to 2D target tracking, 3D target tracking can provide richer information.
Ren Chunlin shared his research experience in non-local multi-view stereo reconstruction. Multi-view stereo matching (MVS) is a fundamental problem in computer vision aimed at restoring the 3D geometric structure of a target from a set of 2D images. Recent advances in the field of MVS have shown that perceiving non-local structural information is crucial for restoring geometric structures in texture-less regions. In the paper accepted at the ICCV23 conference, Ren Chunlin and team members proposed a hierarchical prior mining (HPM-MVS) method for non- local multi-view stereo matching. The experimental results on ETH3D and Tanks&Temples also confirmed that this method has excellent performance and strong generalization ability.
The SCS would continue to serve the development of students and actively create an academic atmosphere that encourages innovation, integration, and mutual learning through the @ World "Touching Academic Frontiers" international exchange series activities, thus providing strong support for the cultivation of composite high-level innovative talents and the delivery of excellent scientific research results. Meanwhile, we also welcome more students to share their experiences and insights.