Sadaf

The rate-distortion-perception (RDP) tradeoff extends classical rate-distortion theory by emphasizing the generation of visually pleasing reconstructions. In the context of online video compression, the system must extract information from incoming frames in real-time and utilize this data to enhance future frame reconstructions. For this scenario, we explore various perception loss functions and examine the differences they produce in the reconstruction of consecutive frames.

Biography:

Sadaf Salehkalaibar is an Assistant Professor in the Department of Computer Science at the University of Manitoba, where her research and teaching focus on learned image and video processing,  statistical learning, information theory, and visual perception models. Before joining the University of Manitoba, she was a Research Associate at the University of Toronto, where she introduced pioneering theoretical approaches to visual learning in video compression. Previously, she served as an Assistant Professor at the University of Tehran and held sabbatical positions at McMaster University, Télécom ParisTech, and the National University of Singapore. Sadaf has contributed to and reviewed for leading conferences and journals in machine learning and information theory, including NeurIPS, ICML, CVPR, and several IEEE Transactions journals.