 |
Professor Zhaoxiang Zhang, PhDCenter for Research on Intelligent Perception and Computing, National Laboratory of Pattern Recognition, Institute of Automation, Chinese Academy of Sciences University of Chinese Academy of Sciences Email: zhaoxiang[dot]zhang[at]ia[dot]ac[dot]cn zhaoxiang[dot]zhang[at]ieee[dot]org |
Research Interests
I would like to give a brief introduction to my recent research in this page. I have also collected the research highlights as posts, which can be accessed from here. Overall, my current specific research interests include: Biologically-Inspired Neural Networks, Human-like Learning, and the applications in Human-oriented Analysis and Scene Understanding, which are introduced in detail as follows.
Biologically-Inspired Neural Networks
I am always interested in get inspirations form the neuroscience and cognitive science to contruct new neural network models. Many mechanisms have been attempted to embed into the neural network modeling, such as the diverse neuron type, the lateral connections of neurons, diverse receptive fields, memory, attention and feedback. Some models have already shown the significant adavantages in the task of object detection, semantic segmentation, et al.
Human-like Learning
I am also interested in advanced machine learning methods. Specifically, I am trying to conquer the disadvantages of the current deep learning based methods, like the requirement of labelled data, lack of explainability, lack of adaptiveness, evolution power in the open environment. Some works have been done in the weakly-supervised learning, domain adaption and continuous learning to mimic the human learning. Some of the methods have achived satisfactory performances in various taks.
Human-oriented Analysis
I am interested in all kinds of human-oriented analysis, like human detection, human tracking, crowd analysis, action recognition, person search, person re-identfication, pose estimation and person generation. Some of the works have achieved satisfactory performances in diverse dataset and benchmarks. We have already implemented some real systems like surveillance, and showed the practicability in real use.
Visual Scene Understanding
I am also interested in the visual scene understanding, specially like 2D object detection, semantic segmentation, instance segmentation, 3D object detection. We have also implemented systems in the scenario of robots and autonomous driving. Owing to the biologically-inspired modeling and human-like learning, our visual scene understanding have achived satisfactory performances in diverse tasks.
