Victor Sanchez
University of Warwick, UK
Can synthetic face images generated by diffusion models be explained?
Abstract. In this talk, I will introduce a state-of-the-art approach designed to explain face images generated by diffusion models. Specifically, I will introduce the Explainable DIffusion PRobabilistic (EDIPR) model, which is based on a classification framework. EDIPR consists of three stages: an initial clustering stage, which serves as the pre-processing step to discover groups of similar face images in the training set; a synthesizing stage, carried out by a diffusion model; and an explaining stage, which allows determining which training images contributed the most to the generation of a new face image. To provide explainability, I will also introduce two influence scores as quantitative metrics: the Normalized Influence Score (NIS) and the class-Normalized Influence Score (cNIS). These scores provide the probability that a specific training image, or class, contributes to the generation of a synthetic face image. Based on synthetic images generated using real images of the FFHQ dataset as training data, I will show that EDIPR provides robust and plausible explanations linking the training images to the synthetic images at three levels of granularity: the region, the image, and the class level.
Bio. Victor Sanchez is a Professor with the Department of Computer Science, University of Warwick, UK, where he leads the Signal and Information Processing (SIP) Lab. He received the PhD degree from The University of British Columbia, Canada, in 2010, and was later a Postdoctoral Researcher at The University of California at Berkeley, USA. His research focuses on the application of signal processing and AI in image and video analysis, biometrics, and security. He has authored over 150 papers in these areas. He was the General Chair of the 11th International Workshop on Biometrics and Forensics (IWBF 2023) and General Co-chair of the 2022 Workshop on Artificial Intelligence for Multimedia Forensics and Disinformation Detection (AI4MFDD 2022) . He has been a member of the Information Forensics and Security Technical Committee of the Institute of Electrical and Electronics Engineers (IEEE) and is currently the Chair of the Technical Committee on Computational Forensics under the auspices of the International Association for Pattern Recognition (IAPR). His research is currently funded by Ford Motor Company USA and the Defense and Security Accelerator of the UK’s Home Office.
Hatef Otroshi Shahreza
Idiap Research Institute, Switzerland
Synthetic Data for Face Recognition
Abstract. State-of-the-art face recognition models are trained on large-scale datasets, collected by crawling the Internet and without individuals' consent, raising legal, ethical, and privacy concerns. Recently, the use of synthetic data to complement or replace real data for the training of face recognition models has appeared as a promising solution. In particular, the recent advancement in generative models provides powerful tools to generate synthetic face images. However, generating face recognition datasets with sufficient inter-class and intra-class variations is still a challenging task. In this talk, I review some recent works on generating synthetic datasets and different approaches
for training face recognition models based on synthetic data. I also discuss the challenges in the existing methods and outline potential future directions.
Bio. Hatef Otroshi Shahreza is a Postdoctoral Researcher at the Biometrics Security and Privacy Group, Idiap Research Institute, Switzerland. He received his PhD from the École Polytechnique Fédérale de Lausanne (EPFL), Switzerland, in 2024, and was a Research Assistant at Idiap Research Institute, where he received Marie Skłodowska-Curie Fellowship for his doctoral program. During his PhD, he also spent six months as a Visiting Scholar with the Biometrics and Internet Security Research Group, Hochschule Darmstadt (HDA), Germany. He is also the winner of the European Association for Biometrics (EAB) Research Award 2023. His research interests include deep learning, computer vision, generative models, and biometrics. He has authored more than 50 papers and holds 3 US patents. He is an Associate Editor of IEEE Transactions on Technology and Society, and has been actively contributing as a reviewer for different conferences and journals (such as NeurIPS, ICLR, ICML, ECCV, etc.).