How Effective Human Body Pose Estimation Methods for Analyzing Complex Dance Movements and Human Gesture?
Alain Tremeau, Université Jean Monnet in Saint Etienne, France
JPEG AI: The First International Standard for Image Coding Based on an End-to-End Learning-Based Approach
João Ascenso, Instituto Superior Técnico, Portugal
A Deep Dive Into Regression
Valerio Giuffrida, University of Nottingham, United Kingdom
How Effective Human Body Pose Estimation Methods for Analyzing Complex Dance Movements and Human Gesture?
Alain Tremeau
Université Jean Monnet in Saint Etienne
France
Brief Bio
Alain Trémeau is Professor at University Jean Monnet and Vice-Rector for International Relations. He is member of the Laboratoire Hubert Curien (UMR 5516). His research activity covers several topics and applications related to color imaging, color science and computer vision. He published numerous book chapters and articles in the field of color science and recently in the domain of cultural heritage. He coordinates two international master degrees in the fields of color science and computer vision (COSI and 3DMT).
Abstract
The relevance and robustness of pose estimation methods is of primary importance in dance movements analysis. The main objective of this keynote paper is to qualitatively compare the most re cent pose estimation models for dance movements analysis and discuss their strengths and weaknesses. For this purpose, we developed a specific methodology and tools. The second objective of this paper is to discuss the interest of human body pose estimation for dance movements analyse and to show that the accuracy of body pose estimation is, in the con text of dance movements analyse, important but less important than the accuracy of the dance movements modelling. Beyond these objectives, we also discuss the lack of efficient models to describe the kinematics of dance movements and human gesture.
JPEG AI: The First International Standard for Image Coding Based on an End-to-End Learning-Based Approach
João Ascenso
Instituto Superior Técnico
Portugal
![]()
Brief Bio
João Ascenso is a professor in the Department of Electrical and Computer Engineering at Instituto Superior Técnico and a member of the Multimedia Signal Processing Group at Instituto de Telecomunicações in Lisbon, Portugal. He earned his E.E., M.Sc., and Ph.D. degrees in Electrical and Computer Engineering from Instituto Superior Técnico in 1999, 2003, and 2010, respectively. He currently serves as the chair of the JPEG CPM (Coding and Performance for Machines) subgroup and the JPEG AI ad-hoc group, where he leads efforts focused on evaluating and developing event-based and learning-based image solutions. With over 150 publications in international journals and conferences, he has accumulated more than 5,000 citations and an h-index of 33. João Ascenso has served as an associate editor for IEEE Transactions on Image Processing, IEEE Signal Processing Letters, and IEEE Transactions on Multimedia. He was the Technical Program Chair for PCS2022 and EUVIP2022 and has contributed to the organizing committees of prominent international conferences, including IEEE ICIP 2023, IEEE ICME 2020, IEEE MMSP 2020, and IEEE ISM 2020. He has received three Best Paper Awards. His research interests include visual coding, quality assessment, 3D visual representation processing, machine coding, super-resolution, and denoising, among others.
Abstract
The JPEG Standardization Committee has recently standardized the JPEG AI standard, marking the introduction of the first image coding specification that utilizes an end-to-end learning-based method. By harnessing cutting-edge deep learning techniques, JPEG AI is designed with future practical applications in mind. The standard has undergone multiple refinements to ensure it is both mature and viable for image encoding and decoding, particularly on mobile devices. When compared to traditional coding systems, JPEG AI offers several distinct advantages: 1) improved rate-distortion performance that enhances perceptual visual quality; 2) considerably faster encoding speeds; and 3) the ability to support diverse optimization goals, such as coding for both human and machine use cases. Built on a learning-based image coding algorithm, JPEG AI generates a compact, single-stream compressed representation that boosts compression efficiency for human visualization, while also delivering strong performance for image processing and computer vision tasks. The goal is to provide a royalty-free baseline for the technology. This talk delves into the core technical principles behind the design of JPEG AI version 1 and presents an outlook on future advancements and extensions of the standard.
A Deep Dive Into Regression
Valerio Giuffrida
University of Nottingham
United Kingdom
Brief Bio
Valerio Giuffrida is an Assistant Professor in Computer Vision at the School of Computer Science, University of Nottingham, and a member of the Computer Vision Lab (CVL). His research focuses on deep learning applied to biological problems, including plant and medical imaging analysis. As a Co-Investigator in the PhenomUK scoping project, Dr. Giuffrida contributes to advancing phenotyping technologies in the UK. He has also organized international workshops on computer vision in plant phenotyping at leading international conferences, fostering collaboration and knowledge exchange in this field.
Abstract
Regression is an important task in machine learning and deep learning, driving a wide range of applications across several real-world applications. From surveillance and agriculture to remote sensing, medical imaging, manufacturing, and transportation, regression is essential for extracting insights and making predictions from data. This keynote will explore the fundamental importance of regression in these fields, with a particular focus on plant phenotyping and medical imaging. I will demonstrate how regression enables precise plant phenotyping, specifically in leaf counting—a key task for monitoring crop health and yield estimation. Through deep learning models trained for regression, we can accurately estimate of the number of leaves in rosette plants, supporting downstream applications in plant science. Transitioning to the medical domain, I will discuss the role of regression in predicting future MRI scans of patients with neurodegenerative conditions. In this context, diffusion models leverage regression to forecast disease progression, offering insights for early diagnosis and treatment planning. Through these real-world applications, this talk will highlight the importance and versatility of regression in modern AI-driven applications and the impact on advancing cross-disciplinary scientific research.