Parallel view-dependent level-of-detail control.

Area of Science:

• Computer Graphics
• Real-time Rendering
• Geometric Modeling

Background:

• Traditional level-of-detail (LOD) control often relies on CPU-bound algorithms.
• Existing GPU approaches struggle with fine-grained mesh adaptation due to parallelism constraints.
• Efficient handling of complex, dynamic scene geometry is crucial for interactive applications.

Purpose of the Study:

• To develop a GPU-implemented scheme for view-dependent level-of-detail (LOD) control.
• To enable selective refinement and coarsening of triangle meshes at the vertex level.
• To achieve highly adapted meshes for dynamic view parameters in real-time.

Main Methods:

• Introduced new data structures and dependency rules for GPU processing.
• Implemented fine-grained progressive mesh updates using parallel streaming passes.
• Overcame GPU's isolated stream element processing challenges.
• Achieved a compact static data structure with minimal memory overhead.

Main Results:

• Demonstrated real-time adaptation of arbitrary triangle meshes on programmable graphics hardware.
• Achieved vertex-level mesh control, outperforming previous CPU-based methods.
• The resulting data structure requires only 57% more memory than an indexed triangle list.
• Enabled real-time exploration of complex models with advanced rendering features.

Conclusions:

• The proposed GPU scheme offers efficient, fine-grained LOD control for dynamic scenes.
• This approach significantly enhances real-time rendering capabilities for complex geometric models.
• The method is suitable for applications requiring dynamic mesh adaptation, including shadowing and transparency.