08:30-09:30 KEYNOTE SPEECH – “Compression factors for realistic 3D graphics” (Video, Slides)
Prof. Marius Preda. Institut MINES-Telecom
Marius Preda received the degree in engineering from the University POLITEHNICA of Bucharest Bucharest, Romania, in 1998, the Ph.D. degree in mathematics and computer science from University Paris V, Paris, France, in 2002, and the eMBA degree from the IMT Business School, Paris, in 2015.
He is currently Associate Professor with the Institut MINES-Télécom, Paris, and also the Convenor of the MPEG 3D Graphics and Haptics Working Group of ISO. He contributed to several international standards with technologies in the fields of 3-D graphics coding, virtual worlds, and augmented reality.
Keynote speech content:
For several decades, the compression of 3D graphics remained a compelling subject of research; however, its tangible impact on practical applications was limited. One predominant reason for this limited applicability was the relatively low resolution of 3D objects, primarily crafted by artists utilizing authoring tools. Recent innovations in capturing 3D scenes and objects, either as point clouds or derivative meshes, have substantially altered this scenario. Such content embodies an extensive data volume, posing challenges for its incorporation into mainstream applications. Yet, the straightforwardness in capturing and rendering their spatial details has escalated their popularity in immersive volumetric data representation.This discourse elucidates contemporary methodologies employed in the compression of realistic 3D graphic content, strategies that have been formulated during the MPEG standardization procedure for the compression of point clouds and dynamic meshes. The first method, termed V-PCC and designed for dense content, transforms the 3D space into 2D patches, leveraging conventional video technologies. In contrast, a second method, G-PCC – designed for sparse point clouds, navigates through the 3D space, generating octree predictors. Empirically, V-PCC attains a compression ratio of 125:1, an object with 1 million points needing a bandwidth of 8 Mbit/s. Concurrently, G-PCC achieves compression ratios up to 10:1 for lossless and 35:1 for lossy configurations. Additionally, a novel strategy focusing on dynamic mesh compression, which involves frame-by-frame simplification and refinements, is delineated. These standardized techniques hold promise in bolstering various domains, such as immersive media, virtual and augmented reality, real-time communication, autonomous navigation, and the preservation of cultural heritage. The talk introduces the perspectives of AI-based graphics compression, an emergent standardization project in MPEG.