Invited Speakers

INVITED TALK 1: Imaging and Metric Considerations for DNNS

Professor Robin Jenkin Ph.D M.Res ASIS FRPS

NVIDIA Corp, Santa Clara, California, USA

Visiting Professor, University of Westminster, UK

Far too often in papers exploring DNN performance, the description of the images used is limited to the pixel count, total number and split between training and validation sets. Not all pixels are created equal for a myriad of reasons. This is further combined with complications due to environmental conditions, such as illumination levels, motion, and variation due to image signal processing. How then is it possible to distinguish between performance limitations imposed by the imaging system and those of the DNN itself? How can we rank cameras intended for our application? Camera-ISP-DNN systems are being deployed extensively within safety critical applications and these questions are becoming increasingly important.

Three existing methods, found in image science literature, are examined to perform ‘first-order’ specification and modeling of imaging systems intended for use in object recognition systems. These are Johnson Criteria, Information Capacity and Detectability. The methods are compared and discussed within the practical context of evaluating modern high dynamic range sensors. Desirable characteristics of quality metrics are considered, and results shown to demonstrate ranking of cameras.

Biography

Robin Jenkin received, BSc(Hons) Photographic and Electronic Imaging Science (1995) and his PhD (2001) in the field of image science from University of Westminster. He also holds a M.Res Computer Vision and Image Processing from University College London (1996). Robin is a Fellow of The Royal Photographic Society, UK, and a board member and VP Publications of Society for Imaging Science and Technology. Robin is secretary of the IEEE P2020 Image Quality for Autonomous Vehicles Standards group and leads the sub-group on image quality for viewing. At NVIDIA Corporation, Robin models image quality for autonomous vehicle applications. He is a Visiting Professor at University of Westminster within the Computer Vision and Imaging Technology Research Group and co-author of the 10th ed. “The Manual of Photography”, Focal Press.

INVITED TALK 2: Impact of Color on Deep Convolutional Neural Networks

Professor Marius Pedersen

Norwegian University of Science and Technology (NTNU)

Deep learning neural networks have become widely used tool for computer vision applications as classification, segmentation and object localization. Recent studies have shown that the quality of images have a significant impact on the performance of deep neural networks. Effects as image noise, image blur, image contrast, and compression artifacts have been studies in relation to the performance of deep neural networks on image classification, but the effects of color have mostly been unexplored. In this talk we study the impact of color distortions in image classification from deep neural networks. Experiments performed using three deep convolutional neural architectures (RESNET, VGGNET and Densenet) show the impact of color distortions on image classification.

Biography

Marius Pedersen is professor at the Norwegian University of Science and Technology. His work is centered on image quality assessment and has more than 70 publications in this field. He received his PhD in color imaging (2011) from the University of Oslo. He is currently the head of the computer science group in Gj¢vik in the department of computer science, and the head of the Norwegian Color and Visual Computing Laboratory, both at NTNU. He is currently Vice President of the US Imaging Science & Technology.