On the evening of November 7, the 12th session of the @World "Touch the Academic Frontier" International Exchange Activity was successfully held at the Corner Conference Room on the second floor of the School of Computer Science. Selected by the China Scholarship Council (CSC) in 2021, two students—Jiao Bingliang and Gao Liying—pursued their studies at Nanyang Technological University, Singapore. They were specially invited to this event to share their research achievements.
Jiao Bingliang’s presentation focused on research into open-world object re-identification technology. As a cutting-edge research direction addressing object recognition and tracking in open environments, this technology aims to tackle the challenges of identifying dynamic, diverse, and even unknown objects in real-world scenarios. It holds great significance in fields such as intelligent surveillance, autonomous driving, and ecology. In practical settings, object re-identification models are hampered by difficulties in data annotation, variability in shooting angles, and the complexity of application environments, which result in insufficient recognition accuracy and generalization capability of the models. To address these issues, Jiao Bingliang introduced several approaches at the presentation, including an unsupervised re-identification algorithm based on post-enhancement of pre-trained models, cross-view feature learning based on weak supervision, and a class-generalized re-identification model based on dual-path knowledge.
Gao Liying shared her research on fine-grained image retrieval methods based on textual information. This technology enables the retrieval of specific images from image databases using natural language descriptions, and plays an important role in public security and economic fields. Nevertheless, the technology currently faces two major challenges: image noise interfering with the extraction of visual target information, and difficulties in cross-modal part alignment. To resolve the problem of image noise interference, Gao Liying introduced two image-text retrieval algorithms—one based on discriminative information extraction guided by visual semantics, and the other based on discriminative information extraction guided by textual knowledge. For the challenge of cross-modal part alignment, she presented an image-text retrieval algorithm that achieves cross-modal part alignment under the guidance of text bias.
