Le prochain séminaire invité IPI aura lieu vendredi 6 juillet de 14h à 15h à Polytech, salle D005.
L’orateur est Pr. Dietmar Saupe, professeur à l’Université de Constance en Allemagne.
Titre du séminaire : » KonIQ-10k: une base de données écologique pour l’apprentissage en profondeur de l’évaluation de la qualité d’image à l’aveugle »
**********
The next IPI seminar will held on Friday the 6th of July (2pm-3pm), in D005 room at Polytech.
The speaker is Pr. Dietmar Saupe, who is full professor at the University of Konstanz, in Germany.
Title of the seminar: « KonIQ-10k: An ecologically valid database for deep learning of blind image quality assessment »
Abstract: This talk is about building a large and diverse image quality database via crowdsourcing, and introducing a deep learning approach that can make best use of it. The main challenge in applying state-of-the-art deep learning methods to predict image quality in-the-wild is the relatively small size of existing quality scored datasets. The reason for the lack of larger datasets is the massive resources required in generating diverse and publishable content. We present a new systematic and scalable approach to create large-scale, authentic and diverse image datasets for Image Quality Assessment (IQA). We show how we built an IQA database, KonIQ-10k, consisting of 10,073 images, on which we performed very large scale crowdsourcing experiments in order to obtain reliable quality ratings from 1,467 crowd workers (1.2 million ratings). We argue for its ecological validity by analyzing the diversity of the dataset, by comparing it to state-of-the-art IQA databases, and by checking the reliability of our user studies. Our novel BIQA method is based on deep learning with convolutional neural networks (CNN); it is trained on full and arbitrarily sized images rather than small image patches or resized input images as usually done in CNNs for image classification and quality assessment. The resolution independence is achieved by pyramid pooling. This work is the first that applies a fine-tuned residual deep learning network (ResNet-101) to BIQA. In contrast to previous methods we do not train to approximate the MOS directly, but rather use the distributions of scores. Experiments were carried out on three benchmark image quality databases. The results showed clear improvements of the accuracy of the estimated MOS values, compared to current state-of-the-art algorithms.
Bio: Since 2002 Dietmar Saupe is a full professor for Computer Science at the University of Konstanz, Germany. He is head of a research group focussing on multimedia signal processing including applications in sports science. Previous positions as professor or lecturer were at the universities of Leipzig (1998-2002), Freiburg (1993-98), Bremen (1987-93), all of them in Germany, and at the University of California, Santa Cruz. U.S.A. (1985-87). His academic degrees are are a diploma, a doctorate and a habilitation degree, all in (applied) mathematics and from the University in Bremen.
Over the years Dietmar Saupe’s areas of interest included numerical methods, dynamical systems, scientific visualization, computer graphics, image and video compression, medical image processing, computer vision, 3D models, and sports informatics. He has co-authored several award-winning books on fractals and chaos. Currently, his research group is engaged in two projects, one on image and video quality assessment and the other on modelling and optimizing performance in endurance sports. He is member of the International Association Computer Science and Sport, and the German professional associations for mathematics and computer science. Saupe currently advises 5 doctorate students and had been academic advisor of 23 more doctorate students, three of whom became professors in computer science or electrical engineering. At the University of Konstanz, he has been Chair of the Department of Computer Science and Vice Dean of the Faculty of Natural Sciences for four years.
