研究目的
To explore the use of a Compound Deep Image (CDI) and raycasting for determining the distance between Virtual Point Lights (VPLs) and geometry for indirect lighting, with a focus on fast building and raycasting performance.
研究成果
The CDI approach outperforms previous work in build time with minimal memory overhead, making it suitable for real-time applications. Raycasting for VPL range determination is efficient, constituting only about 1% of total frametime, and scales well with thousands to millions of VPLs. Future work could integrate this with other VPL techniques for improved visual quality.
研究不足
The approach is slower than assigning random light sizes, with raycasting being approximately 10 times slower. It focuses on performance rather than visual quality comparisons with previous techniques. The method may not be suitable for all dynamic scenes due to precomputation aspects, and shadows from primary light sources are handled separately.
1：Experimental Design and Method Selection:
The methodology involves building a CDI using linearised arrays for memory coherence, raycasting through the CDI to determine VPL ranges, and applying this to indirect lighting. Theoretical models include spherical coordinates for direction sampling and GPU-based sorting techniques.
2：Sample Selection and Data Sources:
Two scenes are used: the Sponza Atrium (approx. 280,000 triangles) and the Rungholt outdoor town scene (approx. 7 million triangles), sourced from McGuire (2017).
3：7).
List of Experimental Equipment and Materials:
3. List of Experimental Equipment and Materials: An NVIDIA GeForce GTX 1060 GPU, driver version 390.25, rendering at HD (1920x1080) resolution. Software includes custom shaders for rasterization and raycasting.
4：25, rendering at HD (1920x1080) resolution. Software includes custom shaders for rasterization and raycasting.
Experimental Procedures and Operational Workflow:
4. Experimental Procedures and Operational Workflow: Steps include capturing and sorting the CDI, rendering g-buffer and RSM, building a VPL buffer with raycasting for range determination, building a light grid, and applying lighting. Raycasting involves casting six rays per VPL.
5：Data Analysis Methods:
Performance is measured in milliseconds for various steps, comparing build times, raycasting times, and total frametime. Statistical analysis includes linear scaling with number of rays.
独家科研数据包，助您复现前沿成果，加速创新突破
获取完整内容
