Skip to main content

LUP Student Papers

LUND UNIVERSITY LIBRARIES

Real-Time Geometry Reconstruction with Compute Shader Tessellation

Yu, Jintao LU (2025) MAMM15 20251
Department of Design Sciences
Abstract
Rendering high-fidelity geometry in real time poses significant challenges due to the large data size, increased computational demands, and memory bandwidth limitations associated with dense meshes. While several techniques have been proposed to address these challenges — such as discrete level-of-detail (LOD) systems, hardware-based tessellation, and virtualized geometry solutions like Nanite. However each comes with its own limitations and constraints which become more pronounced when adapting to modern rendering demands such as dynamic displacement, animated meshes, flexible LOD strategies, and memory-aware mesh reconstruction. To overcome these limitations, we introduce a compute-shader-based geometry reconstruction framework that... (More)
Rendering high-fidelity geometry in real time poses significant challenges due to the large data size, increased computational demands, and memory bandwidth limitations associated with dense meshes. While several techniques have been proposed to address these challenges — such as discrete level-of-detail (LOD) systems, hardware-based tessellation, and virtualized geometry solutions like Nanite. However each comes with its own limitations and constraints which become more pronounced when adapting to modern rendering demands such as dynamic displacement, animated meshes, flexible LOD strategies, and memory-aware mesh reconstruction. To overcome these limitations, we introduce a compute-shader-based geometry reconstruction framework that leverages precomputed tessellation patterns to dynamically refine coarse meshes in real time.
The proposed system is fully programmable, supports arbitrary topologies, and enables GPU-side generation of high-density primitives with efficient memory usage. Through experiments, we compare our method against widely adopted approaches. We demonstrate that our method reduces mesh loading times and improves rendering efficiency while maintaining comparable visual quality. This work contributes a practical, scalable tessellation solution, and offers a step toward flexible, shader-driven geometry pipelines suited for future real-time graphics applications. (Less)
Please use this url to cite or link to this publication:
author
Yu, Jintao LU
supervisor
organization
course
MAMM15 20251
year
type
H2 - Master's Degree (Two Years)
subject
keywords
Computer Graphics, Geometry Processing, GPU-based Tessellation, Mesh Refinement
language
English
id
9198096
date added to LUP
2025-06-13 12:21:15
date last changed
2025-06-13 12:21:15
@misc{9198096,
  abstract     = {{Rendering high-fidelity geometry in real time poses significant challenges due to the large data size, increased computational demands, and memory bandwidth limitations associated with dense meshes. While several techniques have been proposed to address these challenges — such as discrete level-of-detail (LOD) systems, hardware-based tessellation, and virtualized geometry solutions like Nanite. However each comes with its own limitations and constraints which become more pronounced when adapting to modern rendering demands such as dynamic displacement, animated meshes, flexible LOD strategies, and memory-aware mesh reconstruction. To overcome these limitations, we introduce a compute-shader-based geometry reconstruction framework that leverages precomputed tessellation patterns to dynamically refine coarse meshes in real time.
The proposed system is fully programmable, supports arbitrary topologies, and enables GPU-side generation of high-density primitives with efficient memory usage. Through experiments, we compare our method against widely adopted approaches. We demonstrate that our method reduces mesh loading times and improves rendering efficiency while maintaining comparable visual quality. This work contributes a practical, scalable tessellation solution, and offers a step toward flexible, shader-driven geometry pipelines suited for future real-time graphics applications.}},
  author       = {{Yu, Jintao}},
  language     = {{eng}},
  note         = {{Student Paper}},
  title        = {{Real-Time Geometry Reconstruction with Compute Shader Tessellation}},
  year         = {{2025}},
}