Her thesis “Human-centric quality assessment and visual attention modeling for point clouds” was written under supervision of CWI’s DIS-group leader Prof. Dr. P. S. Cesar Garcia and Dr. I. Viola. As the digital world moves toward more immersive experiences - such as videoconferencing and 3D remote communication - the demand for high-quality volumetric video is skyrocketing. A single second of high-quality volumetric video can contain millions of points in space. This makes 3D content very data-heavy. The dataset is too heavy to move from a server to a headset in real-time and has to be compressed and to be streamed efficiently. The question is how can we do this without the user noticing the shortcut?
Xuemei Zhou’s s research bridges the gap between raw point-cloud data and how human beings actually perceive and interact with it. During her PhD, Zhou developed several objective quality metrics grounded in human perception, including M3-Unity and Visam-PCQA. Her work has already gained international acclaim; her PointPCA+ model secured second place in the 2025 Point Cloud Quality Assessment Grand Challenge at the International Conference on Image Processing (ICIP) and will have practical implications for eXtended Reality (XR), telepresence, and remote communication.
Groundbreaking Metrics and Global Recognition
To better understand where users look when wearing a VR headset, she created two significant datasets: QAVA-DPC and TF-DPC, about which she wrote a paper as contribution to IEEE VR 2025 conference. These tools allow researchers to model visual attention and investigate how different tasks (like watching a performance versus watch a performance and assessing the visual quality of the performance) change viewing behavior. By integrating human perceptual factors into algorithms for compression and rendering, this work enables systems to prioritize the parts of a 3D scene that users actually notice. This leads to better bandwidth efficiency and a more seamless experience for the end-user.
Advancing Open Science and Standardization
Based on user questionnaires and interviews, she further proposed a systematic evaluation framework that combines subjective user studies with objective quality metrics. A hallmark of this PhD research is its commitment to open science and the global research community. Zhou actively contributed to international standardization efforts within bodies like MPEG, and ITU, ensuring her findings help shape the future of global media formats.
Her work also fostered deep international ties, collaborating with researchers from the University of Amsterdam, Tampere University, Concordia University, National Tsing Hua University, and George Mason University. Together with CWI and Ultra Video Group (UVG) at Tampere University, she co-created the UVG–CWI–DQPC dataset for real-world volumetric video applications, which was presented in a paper for ACM Multimedia 2025. It is provides a unique benchmark for mixing professional-grade data with everyday 3D captures. This will help the research community improve the quality of 3D video for everyone.
Her PhD was supported through the NWO Women In Science Excel (WISE) grant and the European Commission Horizon Europe project TRANSMIXR (grant agreement 101070109).
About the thesis
· Thesis Title: Human-centric quality assessment and visual attention modeling for point clouds
· Promotor: Prof. Dr. P. S. Cesar Garcia (TU Delft & CWI) Co-promotor: Dr. I. Viola (CWI)
· Date: Wednesday, 4 March 2026; 10:00
· Location: Senate Hall of the Aula Conference Centre, Mekelweg 5, Delft
Relevant previous news:
· https://www.cwi.nl/en/news/dis-group-at-point-cloud-visual-quality-assessment-grand-challenge/
· Full paper of Xuemei Zhou at IEEE VR 2025: “Comparison of Visual Saliency for Dynamic Point Cloud: Task-free vs. Task-dependent”
