|
Hi, I am Hefei Mei (梅鹤飞), a second year CS PhD Student studying at the City University of Hong Kong, supervised by Prof. Minjing Dong. Before that, I received B.S and M.S in Information and Communication Engineering at University of Electronic Science and Technology of China, supervised by Prof. Hongliang Li. My research interest lies in Computer Vision, Trustworthy AI, etc. Currently, I am exploring adversarial robustness of Large Vision Language Models. Please feel free to drop me an Email for any form of communication or collaboration! Email / Google Scholar / Github / LinkedIn / WeChat / Red Notes (小红书） |
|
|
|
|
|
Hefei Mei, Zirui Wang, Shen You, Minjing Dong, Chang Xu Preprint While existing effective attacks always focus on task-specific white-box settings, these approaches are limited in the context of LVLMs, which are designed for diverse downstream tasks and require expensive full-model gradient computations. Motivated by the pivotal role and wide adoption of the vision encoder in LVLMs, we propose a simple yet effective Vision Encoder Attack (VEAttack), which targets the vision encoder of LVLMs only. |
|
Hefei Mei, Minjing Dong, Chang Xu AAAI Conference on Artificial Intelligence (AAAI-25) We introduce an efficient Image-to-Image diffusion classifier with a pruned U-Net structure and reduced diffusion timesteps. Besides the framework, we redesign the optimization objective of diffusion models to fit the target of image classification, where a new classification loss is incorporated in the DM-based image translation framework to distinguish the generated label from those of other classes. We conduct sufficient evaluations of the proposed classifier under various attacks on popular benchmarks. |
|
Hefei Mei, Taijin Zhao, Shiyuan Tang, Heqian Qiu, Lanxiao Wang, Minjian Zhang, Fanman Meng, Hongliang Li Neurocomputing We consider the conversion relationship between non-similar samples of the same category (Local Reverse Samples, LRSamples) in the base class can be learned and also transferred to the novel class. So we propose the selection rule and intra-class feature converter (IFC) of LRSamples in our Center Calibration Variance Augmentation (CCVA) module to adaptively adjust the center position of the novel class distribution for FSOD. Moreover, we propose a Feature Density Boundary Optimization (FDBO) module to adaptively adjust the importance of samples depending on their distance from the decision boundary. |
|
Hefei Mei, Hongliang Li, Heqian Qiu, Jianhua Cui, Longrong Yang IEEE International Conference on Visual Communications and Image Processing (VCIP-22) We present a new feature extraction method, DE-Crossline, which can enhance the original crossline representation to capture more accurate object information. Specifically, we divide the crossline into several segments, each of which extracts the maximum activation key point respectively to reduce the impact of noise mentioned above. Furthermore, considering various shapes and sizes of objects, we design a Deformable Width Extension Module to learn a suitable width of each crossline, so as to capture richer object information. |
|
|
 |
City University of Hong Kong
Hong Kong 2024.09 - present Ph.D. in Computer Science GPA: 4.15 / 4.30 Supervisor: Prof. Minjing Dong |
 |
University of Electronic Science and Technology of China
Chengdu, China 2021.09 - 2024.06 M.S. in Information and Communication Engineering GPA: 3.88 / 4.00 Supervisor: Prof. Hongliang Li |
|
University of Electronic Science and Technology of China
Chengdu, China 2017.09 - 2021.06 B.S. in Electronic and Information Engineering GPA: 3.92 / 4.00 |
|
|
|
|
|
Template courtesy: Jon Barron. |