Ruhan Conceição & Luciano Agostini
Universidade Federal de Pelotas (Brazil)
Neural video coding: when machines learn to compress
Language: Portuguese
Date: October 7th, 10h00
Abstract
We will provide an overview of neural video coding (NVC), a class of video compression systems based on deep learning models trained end-to-end to minimize rate-distortion loss. Unlike traditional codecs that rely on hand-crafted tools, NVC systems learn to optimize motion estimation and compensation, residual coding, and entropy modeling directly from data. In addition to improving rate-distortion performance, NVC enables new coding paradigms through contextual and conditional modeling, allowing for content-adaptive representations that surpass the capabilities of conventional approaches. In this talk, we will review fundamental concepts such as convolutional neural networks, autoencoders, and variational autoencoders. We will also briefly present recent academic models that have demonstrated superior performance compared to traditional state-of-the-art codecs. Finally, we will address ongoing challenges, including encoder-decoder complexity, hardware limitations, and cross-platform compatibility.
Biography
Ruhan Avila da Conceição holds a Master’s degree in Computer Science (2015) and a Bachelor’s degree in Computer Engineering (2014) from the Federal University of Pelotas, where he is currently pursuing a Ph.D. in Computer Science. In 2024, he was a visiting Ph.D. researcher at National Yang Ming Chiao Tung University in Taiwan. He is a Professor at the Federal Institute of Education, Science, and Technology of Sul-rio-grandense, where he coordinated the Integrated Technical Program in Internet Informatics from 2019 to 2022. He completed the Deep Learning Specialization by DeepLearning.AI, strengthening his expertise in neural networks and modern AI techniques. His current research interests include neural video compression, deep learning, video coding, and hardware architectures for visual signal processing. With over a decade of experience in academia and R&D, he has authored several programming books and published extensively in leading journals and international conferences. He also serves as a reviewer for top-tier journals and conferences such as JETCAS, JRTIP, JICS, Cluster Computing, ISCAS, and PCS.
Luciano Agostini is a Brazilian Distinguished Researcher through a CNPq PQ-1D grant. He received the M.S. and Ph.D. degrees from Federal University of Rio Grande do Sul, Porto Alegre, Brazil, in 2002 and 2007 respectively. He is an Associate Professor since 2002 at Federal University of Pelotas (UFPel), Brazil, where he leads the Video Technology Research Group (ViTech) and the Group of Architectures and Integrated Circuits (GACI). He is advisor at the UFPel Master and Doctorate in Computer Science courses. He was the Executive Vice President for Research and Graduate Studies of UFPel from 2013 to 2017. He has more than 300 published papers in journals and conference proceedings. His research interests include 2D and 3D video coding, algorithmic optimization, arithmetic circuits, and digital systems. Currently he is an Associated Editor of two IEEE CAS journals: TCSVT and OJCAS. Dr. Agostini is a Senior Member of IEEE and ACM. He is a member of the IEEE CAS, CS, and SPS societies. At the IEEE CAS, he is a member of the MSA and VSPC Technical Committees. He is also member of SBC and SBMicro Brazilian societies.