Jiang Liu

I am a PhD student (2022 - now) in the Human-Centric Machine Vision and Intelligence Group within the School of Computer Science and Informatics, Cardiff University, UK, advised by Prof. Hantao Liu and Dr. Katarzyna Stawarz.

Before that, I obtained my M.S. and B.S. degrees from the School of Information and Control Engineering at China University of Mining and Technology in 2022 and 2019, respectively, advised by Prof. Shiyin Li and Dr. Yuan Yang. I used to be a visiting student at Arizona State University, USA (2019) and a research assistant at Southeast University, China (2020).

My research interests primarily lie in image quality assessment, action quality assessment and saliency prediction. I also have a research background in smart buildings and wireless sensing.

          CV                               ORCID          

(* denotes equal contribution, # denotes corresponding author)

Highlight:
• We propose a novel database that integrates chest X-ray images, ground-truth diagnostic decisions, gold standard annotations and gaze data collected from radiologists.
• We propose leveraging machine learning to predict radiologist expertise based on gaze data.

Highlight:
• We propose a novel adaptive spatiotemporal graph module for AQA.
• We propose a novel spatiotemporal graph transformer framework.
• Our results demonstrate that the proposed method achieves state-of-the-art performance on popular AQA datasets.

Highlight:
• The first systematic literature review to investigate up-to-date research in vision-based AQA
• This study offers a detailed examination of 96 papers, including their applications, datasets, data modalities, methods, and evaluation metrics.
• This review identifies current challenges in existing research, providing valuable insights and recommendations for future studies. These suggestions are intended to inspire the development of new methods and applications within the AQA field.

Highlight:
• We devise a novel Adaptive Graph Attention module for deep learning-based IQA.
• We propose a Patch-wise-based Hierarchical Perceptual regression module to combine MSE and deep ordinal (DO) regression for inferring scores from different patches at various depths of the network.
• We show the substantial superiority of the proposed BIQA model over existing alternative models, through extensive experiments on many benchmark datasets.

Highlight:
• Distinguish from stationary monitoring, this study provides a set of schemes that enable a mobile robot with real-time position tracking and IAQ online sensing.
• Mobile sensing provides a self-controlled, high resolution, wireless and trackable strategy with agile adaptions to the dynamic indoor environment.
• Automatically detecting indoor pollutant sources and locating where they are, can be beneficial for environment analysis, building security and energy control.

based on a template by Jonathan Barron