Yuang Shi (施宇昂)

I'm a third year Ph.D. Candidate at National University of Singapore advised by Prof. Wei Tsang Ooi, where I work on Networking and Multimedia Systems, specifically on 3D Media Streaming.

I received my M.Comp Degree in 2022 from National University of Singapore and B. Eng. Degree in 2021 from Sichuan University in Sichuan.

I am a part-time cat person and a full-time dog person.

Email  /  Google Scholar  /  Github

• [2025/01] - Our paper "LTS: A DASH Streaming System for Dynamic Multi-Layer 3D Gaussian Splatting Scenes" is accepted to MMSys 2025.
• [2024/11] - Our paper "LapisGS: Layered Progressive 3D Gaussian Splatting for Adaptive Streaming" is accepted to 3DV 2025.
• [2024/08] - Our paper received the Best Paper Award at the SIGCOMM 2024 Workshop on Emerging Multimedia Systems (EMS 2024)! See the Press from NUS.
• [2024/05] - Invited talk at Sichuan University, Sichuan.
• [2024/04] - Invited talk at Université de Toulouse, Toulouse INP-ENSEEIHT, IRIT, France.

As a PhD student, I mainly focus on 3D Media Streaming, including compression, networking, and quality evaluation.

Some works are highlighted. Authors marked with ^* are interns or students whom I mentored when the work was carried out.

We develop, implement, and evaluate the very first DASH-based dynamic 3DGS streaming system, named Tile, Segment, and Layer Adaptive streaming (LTS). LTS is built on our previous work LapisGS, and achieves superior performance in both live streaming and on-demand streaming.

We introduce LapisGS, a layered progressive 3D Gaussian Splatting (3DGS), which offers a progressive representation supporting a continuous rendering quality adapted for bandwidth-aware streaming.

Inspired by traditional painting techniques, we propose a novel hybrid representation for 3DGS that categorizes Gaussians into (i) Sketch Gaussians, which define scene boundaries, and (ii) Patch Gaussians, which cover smooth regions. This hybrid categorization is conducive to efficient compression, and progressive and scalable streaming.

We propose GSVC, an approach to learning a set of 2D Gaussian splats that can effectively represent and compress video frames. Experiment results show that GSVC achieves good rate-distortion trade-offs, comparable to state-of-the-art video codecs such as AV1 and HEVC, and a rendering speed of 1500 fps for a 1920x1080 video.

We proposed two algorithms, MGA and MGAA, to allocate bitrate across multiple 3DGS scenes for streaming over dynamic networks.

We present QV4, a Quality-of-Experience (QoE) based streaming system for viewpoint-aware V-PCC-encoded volumetric video. QV4 achieves average compression ratio of 610 while keeping satisfactory quality, which is around 80x better than other SOTA streaming systems.

We represent 3D dynamic content as a sequence of NeRFs, converting the explicit representation to neural representation. We then compress the neural representation based on the insight of significant similarity between successive NeRFs.

We propose a QoE model to predict the subjective quality with respect to the compression level of geometry and texture, quantifying the impact of geometry and texture compression on perceptual quality.

We investigated the influence of rendering face quality of the avatars on users' viewing experience in MPEG V-PCC-encoded volumetric videos, and revealed the significant role of facial quality in influencing users' overall perceptual quality in volumetric videos.

We show that it is possible to improve the quality of V-PCC encoded point clouds at low bit-rate by exploiting redundant information among the points in the 3D domain.

When I was a Master student, my dissertation is about human activity recognition in-the-wild.

ShapeCNF is a simple, fast, and accurate change-point detection method which uses shape-based features to model the patterns and a conditional neural field to model the temporal correlations among the time regions.

During my undergraduate, I spent most of my time on medical image analysis.

We propose a consistency training framework for semi-supervised NPC segmentation, which includes an Uncertainty-weighted Prediction Consistency Training (UPCT) strategy and a Relation-driven Consistency Training (RCT) strategy.

ASMFS is a novel multi-modal feature selection method for classification of Alzheimer's Disease, which performs adaptive similarity learning and feature selection simultaneously.

• [2025/01-2025/05] - Graduate Tutor & Teaching Assistant, CS2106 Introduction to Operating Systems.
  • Over 600 students enrolled.
  • Conducted weekly tutorials.
• [2024/08-2024/12] - Graduate Tutor & Teaching Assistant, CS3244 Machine Learning.
  • Over 100 students enrolled.
  • Conducted weekly tutorials; Graded assignments; Mentored course projects; Made Midterm and Final Exam.
• [2024/01-2024/05] - Graduate Tutor & Teaching Assistant, CS3244 Machine Learning.
  • Over 100 students enrolled.
  • Conducted weekly tutorials; Graded assignments; Mentored course projects; Made Midterm and Final Exam.
• [2023/01-2023/05] - Graduate Tutor & Teaching Assistant, CS3244 Machine Learning.
  • Over 100 students enrolled.
  • Conducted weekly tutorials; Graded assignments; Mentored course projects; Made Midterm and Final Exam.
• [2022/08-2022/12] - Graduate Tutor & Teaching Assistant, CS3244 Machine Learning.
  • Over 200 students enrolled.
  • Conducted weekly tutorials; Graded assignments; Mentored course projects; Made Midterm and Final Exam.

I really appreciate the opportunity to work with the following talent students (I only list the selected ones who I have worked with for a long time, and I have made non-negligible contributions to their research projects 😃.):

1. Jia Yi Lee.
   • FYP 2025. Efficient Video Representation with 3D Gaussian Splatting.
2. Longan Wang --> PhD @ NUS.
   • Undergraduate NGNE 2025. Gaussian-based Video Compression.
   • Co-author of an MMSys'25 paper.
3. Chenggang Yang --> Tencent.
   • Master Dissertation 2024. 3D Line Segment Representation on 3D Gaussian Splatting.
   • Co-author of an MMSys'25 paper.
4. Sherwin Poh --> Apple.
   • FYP 2024. Real-time Point Cloud Super-Resolution for Volumetric Video Streaming.
   • Contributor of VVTk, an open-sourced toolkit for volumetric video streaming.
5. Chi Sern Ng --> Tiktok.
   • FYP 2024. Scalable Approaches to Volumetric Video Playback.
   • Contributor of VVTk, an open-sourced toolkit for volumetric video streaming.
6. Joseph Wang Guanlin --> RA @ NUS.
   • FYP 2024, UROP 2023. Improving the Performance of V-PCC through Segmentation With Meta's SAM.
   • Contributor of VVTk, an open-sourced toolkit for volumetric video streaming.
7. Bennett Clement --> HoYoVerse.
   • FYP 2023. Viewpoint-aware VPCC-Encoded Volumetric Video Streaming.
   • Co-author of an MMSys'24 paper.
8. Ruoyu Zhao.
   • Undergraduate Research Intern 2023. Neural-based Representation for Volumetric Video Compression.
   • Co-author of an MMSys'24 paper.
9. Shixin Ji --> PhD @ Brown University.
   • Undergraduate NGNE 2023. Human Activity Recognition In-the-Wild Using Wearable Devices.

• [2024/05] - Invited talk at Sichuan University, Sichuan.
• [2024/04] - Invited talk at Université de Toulouse, Toulouse INP-ENSEEIHT, IRIT, Toulouse.
• [2023/12] - Invited talk at National Tsing Hua University, HsinChu.
• [2022/11] - Student Volunteer. The 21st IEEE International Symposium on Mixed and Augmented Reality, 2022, Singapore.

• IEEE Multimedia.
• Medical Image Analysis (MIA).
• IEEE Transactions on Cognitive and Developmental Systems (TCDS).
• ACM Transactions on Multimedia Computing, Communications, and Applications (TOMM).
• ACM Multimedia (ACM MM) 2024.