Professor at the University of Amsterdam, the Netherlands

Bio

Theo Gevers is a professor of computer vision at the University of Amsterdam. He is director of the Computer Vision Lab and co-director of the Atlas Lab (UvA-TomTom) and Delta Lab (UvA-Bosch) in Amsterdam. His research area is artificial intelligence with the focus on computer vision and deep learning, and in particular image processing, 3D (object) understanding and human-behavior analysis with industrial and societal applications. He is the co-founder of 3DUniversum, Scanm B.V. (Sold in 2020), and Sightcorp (Sold in 2022).

Abstract for presentation

Coloring (generative) AI

 

Professor at Georgia Institute of Technology, US

Bio

Prof. AlRegib is currently the John and Marilu McCarty Chair Professor in the School of Electrical and Computer Engineering at the Georgia Institute of Technology. His group is the Omni Lab for Intelligent Visual Engineering and Science (OLIVES) at Georgia Tech. In 2012, he was named the Director of Georgia Tech’s Center for Energy and Geo Processing (CeGP). He is the director of the Center for Signal and Information Processing (CSIP). He also served as the Director of Georgia Tech’s Initiatives and Programs in MENA between 2015 and 2018. He has authored and co-authored more than 300 articles in international journals and conference proceedings. He has been issued several U.S. patents and invention disclosures. He is a Fellow of the IEEE.

Prof. AlRegib received the ECE Outstanding Graduate Teaching Award in 2001 and both the CSIP Research and the CSIP Service Awards in 2003. In 2008, he received the ECE Outstanding Junior Faculty Member Award. In 2017, he received the 2017 Denning Faculty Award for Global Engagement. Prof. AlRegib has provided services and consultation to several firms, companies, and international educational and R&D organizations. He has been a witness expert in several patents’ infringement cases.

His research group is working on projects related to machine learning, image and video processing and understanding, subsurface imaging, perception, and healthcare intelligence. The primary applications of the research span from Autonomous Vehicles to Portable AI-based Ophthalmology and Eye Exam and from Microscopic Imaging to Seismic Interpretation. The group has produced a number of large-scale datasets; examples can be found at https://alregib.ece.gatech.edu/.

 

Abstract for presentation

Visual Explanations in AI

Visual explanations have traditionally acted as rationales used to justify the decisions made by machine learning systems. With the advent of large-scale neural networks, the role of visual explanations has been to shed interpretability on opaque models. We view this role as the process for the network to answer the question `Why P?’, where P is a trained network’s prediction. Recently however, with increasingly capable models, the role of explainability has expanded. Neural networks are asked to justify `What if?’ counterfactual and `Why P, rather than Q?’ contrastive question modalities that the network did not explicitly train to answer. This allows explanations to act as reasons to make further prediction. The talk provides a principled and rational overview of Explainability within machine learning and justifies them as reasons to make decisions. Such a reasoning framework allows for robust machine learning as well as trustworthy AI to be accepted in everyday lives. Applications like robust recognition, image quality assessment, visual saliency, anomaly detection, out-of-distribution detection, adversarial image detection, seismic interpretation, semantic segmentation, introspection, and machine teaching among others will be briefly discussed.

AI Scientific Director at DXOMARK

Bio

Sira Ferradans is currently the AI director at DXOMARK leading the machine learning team on the application of  computer vision on  precise image quality assessment on natural scenes.  She has worked as a researcher at Duke University (North Carolina, US) and Ecole Normale Superieur (ENS Paris, France) and was keynote speaker at CIC’23. The DXOMARK team has recently focused on Portrait quality assessment, publishing a new Portrait database and organizing a CVPR challenge on the same topic.

Abstract for presentation

Studying user preferences for diverse skin tone portrait quality rendition

Portraits are the most common use case for smartphone photography, however, producing a realistic and pleasant skin tone in real scenarios is still challenging for all manufacturers, especially in common conditions such as night or low light scenes. However, producing non-homogeneous quality rendition across skin tones has become a sensitive issue, and its evaluation is crucial for the industry. In the scientific literature, we find mostly studies that evaluate synthetic modifications of laboratory portraits.  In this talk, we will show the challenges of systematically evaluating diverse skin tones in the lab using realistic mannequins. However, we will also show that real setups are much more complex to evaluate, and user preferences depend on many factors.

We will go through the conclusions obtained during DXOMARK’s last user studies, where we examine the performance of high-end smart-phone cameras in common every-day use cases. This study shows that around 20% of portraits are currently discarded due to quality problems, implying that contemporary smartphone cameras are far from solving the skin tone rendition problem.

These challenges are mostly because there is no clear target definition of user preferences regarding color skin tone rendering. The definition of this target could path the way to automatizing skin tone rendition evaluation with Machine Learning.

 

Charles Poynton

Bio

Charles Poynton is an imaging and colour scientist working in video/HD/UHD/4K/HDR/HDR/D‑cinema. Thirty years ago, he decided upon the count of 1080 image rows for HD (and thereby, “square pixels”). He contributed to the development of color spaces such as BT.709/BT.1886, DCI P3, ACES AP0/AP1, and BT.2020; he was responsible for the introduction of the Adobe RGB colourspace. He received his PhD in 2018 with a thesis entitled, “Colour Appearance Issues in Digital Video, HD/UHD, and D‑cinema.

Abstract for presentation

Technological Natural Selection in Imaging Standards 

Video signal decoding by a CRT’s inherent power function (“gamma”) very nearly inverts the perceptual uniformity of CIE L*. I used to consider this to be an amazing coincidence. In about 1992, I was chatting to Mike Schuster (of Adobe) about CRT gamma, and I commented to him about what I saw as the fluke by which halftone dot gain in printing also has nonlinear behaviour favourable to perception. Michael told me that he had thought about that for a long time. He said that he had reached the conclusion that it was a kind of technological natural selection – if not for optical dot gain, 8-bit CMYK halftoning would have failed, and some other scheme would have eventually been found.

In this talk, I’ll describe several situations in digital colour imaging where suitable – even near-optimum – solutions to problems were found by processes involving mutation and selection pressure, rather than by explicit engineering. There are lessons for imaging system design.