Stamatios Georgoulis is a 3rd-year post-doctoral researcher at the CVL group of ETH Zurich, working with Prof. Luc Van Gool on the topics of image generation, dense prediction, and multi-task learning. In parallel, he holds a computer vision and machine learning researcher position at Huawei's Zurich Research Center, working with Prof. Davide Scaramuzza on computational photography. Before coming to Zurich, he was a doctoral student at the PSI-VISICS group of KU Leuven, where he received his PhD under the supervision of Prof. Luc Van Gool and Prof. Tinne Tuytelaars, also collaborating with Prof. Mario Fritz, Prof. Tobias Ritschel and Dr. Konstantinos Rematas. During his PhD, his research was mainly focused on extracting surface characteristics and lighting - in particular, 3D shape, surface reflectance, and environmental illumination - from images. Further back, before even arriving in Leuven, he received his diploma in Electrical and Computer Engineering from the Aristotle University of Thessaloniki. He conducted the research for his diploma thesis on "Adaptive Pain Detection via Webcam using Advanced Image Processing Techniques" in collaboration with Dr. Stefanos Eleftheriadis under the supervision of Prof. Leontios Hadjileontiadis. During his stay in Thessaloniki, he worked as a student research assistant in the Signal Processing and Biomedical Technology Unit. He regularly serves as a reviewer in major computer vision and machine learning conferences and journals with distinctions, and he has won the best paper award at the IEEE Computer Society Biometrics Workshop of the CVPR 2020 conference.
Deep Learning for Intrinsic Image Decomposition, Multi-Task Learning and Unsupervised Learning
Since the seminal work of ImageNet, neural networks have become the "go-to" approach for many research fields, including computer vision, medical image analysis, natural language processing, and so on. This tremendous success can largely be attributed to a combination of deeper architectures, larger datasets, and better processing units. In this talk, we focus on deep learning techniques applied to a broad range of computer vision applications, including intrinsic image decomposition, multi-task learning and unsupervised learning. We present a short literature review on each topic and analyze good practices to achieve state-of-the-art performance for these tasks within the deep learning framework.