Multi-modal scene understanding

GTA funded PhD studentship in Computing

Project Highlights

1. Proposing balanced learning techniques to mitigate the generic biased prediction issue.
2. Improving the holistic inference capabilities of the multi-modal scene understanding systems.
3. Interpreting the ever-increasing visual data accurately and efficiently.

Project

In the current information age, interpreting the ever-increasing visual data generated by social media or other sources is urgently needed across many related sectors, e.g., robotics, security, or the creative industries. Multi-modal scene understanding research is the key within the above interpretation tasks, which is essentially a critical aspect of computer vision that involves not only identifying objects in a scene but also understanding their relationships. However, such research is still far from desired due to various challenges, such as biased prediction, inefficiency, and inferior inference capabilities.

 One primary aim of this PhD project is to solve the above challenges by proposing advanced deep learning-based multi-modal scene understanding paradigms. These paradigms will improve the holistic inference capabilities as well as the overall efficiencies of the multi-modal scene understanding systems. Moreover, the generic biased prediction issues caused by the imbalanced distributions of the training datasets will also be mitigated.

 The applications of this project are broad and varied, e.g., object detection, scene graph generation, image captioning, or visual question answering. The overarching aim is to interpret the ever-increasing visual data so that the tremendous semantic information embedded within the above visual data can be properly exploited to meet urgent needs across various related sectors.

PhD start date 23 September 2024

Enquiries to project supervisor Dr Daqi Liu dl336@leicester.ac.uk  or cmspgr@le.ac.uk

Further details and application advice at https://le.ac.uk/study/research-degrees/funded-opportunities/cms-gta