Headline One:Introduction to Artificial Intelligence with Deep Learning

Lecturer:Ajmal Saeed Mian

Host:Dr. Junying Chen

Time:14:30-16:00,November 6th,2019

Venue:Academic Report Hall in B8 Building, Southern Campus of SCUT

Abstract:

      Deep learning has achieved breakthough performances for many tasks often surpassing humans. In this talk, I will introduce the basics of deep learning starting from Multi Layer Perceptron and finishing on a few famous Convolutional Neural Network (CNN) architectures. Along the way, I will explain the main concepts of neural networks, what they actually learn, where their power comes form, how they avoid local minima, why greater depth is better and how to design very deep network architectures while avoiding the vanishing gradient problem. This talk assumes no prior knowledge of deep learning and is hence suitable for students at all stages of Engineering.

Headline Two:Deformable 3D face modeling for deep 3D face recognition and medical applications

Lecturer:Ajmal Saeed Mian

Host:Dr. Junying Chen

Time: 14:30-16:00,November 7th,2019

Venue:Academic Report Hall in B8 Building, Southern Campus of SCUT

Abstract:

        In this talk, I will present our research on dense 3D face correspondence, a core problem for many applications such as biometric identification, symptomatology for the diagnosis of Autism and Obstructive Sleep Apnoea and planning for facial reconstructive surgery. From a morphometric point of view, we are interested in performing dense correspondence based purely on shape without using texture. This makes the problem challenging but the correspondences and subsequent analyses more precise. The idea is to start from a sparse set of automatically detected corresponding landmarks and propagate along the geodesics connecting nearby points. By anchoring on the most reliable correspondences for propagation, accurate dense correspondences are iteratively established between hundreds of faces without using a prior model. Thus, we are able to construct population specific deformable face models for symptomatology and patient specific morphs to facial norms for reconstructive surgery. Moreover, by establishing dense correspondences between different facial identities and expressions, we synthesize millions of 3D faces with varying identities, expressions and poses to learn a deep Convolutional Neural Network (FR3DNet) for large scale 3D face recognition. FR3DNet achieves state-of-the-art results, outperforming existing methods in open-world and close-world face recognition, on a dataset four times the largest dataset reported in the existing literature. At the end, I will show how our methods are used to diagnose Obstructive Sleep Apnea, Autism Spectrum Disorder in children and for orthodontic surgery planning/analysis.

Headline Three:Precision Modeling of 3D Human Motion: Behaviour and Performance Analysis

Lecturer:Prof. Ajmal Saeed Mian

Host:Dr. Junying Chen

Time: 14:30-16:00, November  8th,2019

Venue:Academic Report Hall in B8 Building, Southern Campus of SCUT

Abstract:

        In this talk, I will present our research on modeling 3D human motion with applications to human action recognition, recovery of 3D human pose from monocular video and athlete performance optimization. At the heart of our methods (published in multiple journals like PAMI, IJCV, TBME, CVPR) is a strategy to avoid tedious data annotation. Instead, we capitalize on legacy motion capture (MoCap) data to synthesize videos by riging (to the MoCap) 3D humans with varying sizes, gender and clothing textures, placing them in random backgrounds and rendered them from 180 camera viewpoints under random lighting. We also model their clothes deformations using a physics engine. Thus, we generate big data with known ground truth for training deep models for human action recognition from skeleton/joint data or videos. We also propose a method for recovering the full mesh 3D human pose from monocular video. Our athlete performance optimization models learn from legacy MoCap data to estimate the ground reaction forces and moments during running and side stepping. These measurements are essential to model knee joint moments and could only be performed inside labs using expensive force plates. Our research potentially brings this capability to the sports fields.

Short bio of Lecturer:

         Ajmal Mian is a Professor of Computer Science at The University of Western Australia. He has received two prestigious fellowships and several research grants from the Australian Research Council and the National Health and Medical Research Council of Australia with a combined funding of over $12 million. He was the West Australian Early Career Scientist of the Year 2012 and has received several awards including the Excellence in Research Supervision Award, EH Thompson Award, ASPIRE Professional Development Award, Vice-chancellors Mid-career Research Award, Outstanding Young Investigator Award, the Australasian Distinguished Doctoral Dissertation Award and various best paper awards. He is an Associate Editor of IEEE Transactions on Image Processing and the Pattern Recognition journal. He has also served or is serving as a Guest Editor for special issues in Remote Sensing, Neural Computing & Applications, PR, CVIU and CVIU. He is a General Chair of the International Conference on Digital Image Computing Techniques and Applications (DICTA) 2019. He was a General Chair of the Asian Conference on Computer Vision 2018, Program Chair of DICTA 2012 and Area Chair of WACV 2019, WACV 2018, ICPR 2016 and ACCV 2014. Ajmal Mian has supervised 13 PhD theses to completion and has published over 180 scientific papers in prestigious journals and conferences including IEEE TPAMI, IEEE TNNLS, IEEE TIP, PR, IEEE TGRS, IEEE TITS, IEEE TBME, CVPR, ICCV and ECCV. His research interests are in computer vision, machine learning including defence against adversarial attacks, 3D shape and point cloud analysis, facial recognition, human action recognition and video analysis.