Towards reliable deployment of computer vision systems in the real world. This project aims to enhance the reliability of computer vision models in real-world deployment by quantitatively assessing th

Description

Towards reliable deployment of computer vision systems in the real world. This project aims to enhance the reliability of computer vision models in real-world deployment by quantitatively assessing the environment, facilitating optimal model selection and enabling accurate expression of prediction uncertainty. Current literature falls shorts in extending model choices beyond the training domain or adjusting uncertainty to varied test conditions, posing safety risks. This project is significant for advancing the theoretical understanding of model performance in complex and changing environments and promote practical applicability. Anticipated outcomes include innovative techniques for improving and signaling model reliability, with substantial benefits to computer vision applications such as autonomous vehicles. . Scheme: ARC Future Fellowships. Field: 4603 - Computer Vision and Multimedia Computation. Lead: A/Prof Liang Zheng