About this Workshop

Recent years have seen significant advances in image processing and computer vision applications based on Deep Neural Networks (DNNs). Often deep neural networks for such applications are trained and validated based on the assumption that the images are artefact-free. However, in most real-time embedded system applications the images input to the networks, in addition to any variations of external conditions, have artefacts introduced by the Image Signal Processing (ISP) pipelines.

Despite recent advances in interpretability and explainability of deep neural models, DNNs remain widely systems whose operational boundaries cannot be explained or otherwise quantified. It is therefore not clear the level of ISP distortions critical networks can tolerate, or the exact reasons for any performance degradation.

This workshop addresses the issues of performance quantification in DNNs and explore recent advances in the systematic analysis of the performance of deep neural networks with respect to degradations in the input image quality due to the ISP pipeline and their proposed solutions.

Topics

Topics of interest include (but are not limited) to:

• Case studies investigating the performance of deep neural networks with respect to change in input image quality

• Operational boundaries of DNNs with respect to input image quality

• Input image quality metrics for deep neural networks

• Optimisation of physical camera parameters and ISP pipelines for integrated DNNs embedded systems

• Architectural structures of DNNs for optimising integrated ISP embedded systems