Practical Volume Rendering for realtime applications
Download as PPTX, PDF•3 likes•4,383 views
The document discusses practical volume rendering techniques for real-time applications, highlighting its use in various fields like medical imaging, gaming, and procedural generation. It covers direct and indirect rendering methods, such as ray-marching and marching cubes, and emphasizes the importance of handling large meshes and level of detail using algorithms like transvoxel. Additionally, it addresses potential challenges such as memory consumption and texturing while offering a comparison of different algorithms.
Practical Volume Rendering for realtime applications
- 2. Practical Volume rendering for realtime applications Stoyan Nikolov
- 3. VIDEO
- 4. Who am I Stoyan Nikolov [email protected] @stoyannk Co-Founder & Tech lead @ Coherent Labs Senior Graphics dev @ Masthead Studios Senior dev @ Eutelia
- 5. Overview Why use Volume rendering at all? What is Volume rendering? Types of Volume rendering techniques Practical (today) rendering techniques
- 6. Why Volume rendering • Allows us to see „inside“ an object • Medical imaging • Industry • Biology
- 7. Why Volume rendering - Games Freedom of form
- 8. Why Volume rendering - Games Destructability EverQuest Next by Sony Online Entertainment
- 9. Why Volume rendering - Games Procedural generation
- 10. Volume rendering - Input Usually discrete volume elements — voxels
- 11. Direct rendering in RT • • • • Ray-marching distance fields Gigavoxel Sparse voxel octrees (SVO) SVO for Global Illumination slisesix demo by rgba Unreal Engine 4 by Epic Games
- 12. Indirect rendering for RT • • • Create a polygon mesh GPUs are designed for polygons Other subsystems usually expect polygons
- 13. Basic algorithms Marching cubes / tetrahedra / MC 33
- 14. Realtime requirements • We want to support very large meshes • Level of Detail (LOD) required • Difficult with MC • Transvoxel designed to solve this
- 15. Transvoxel • Outline in Eric Lengyel’s PhD thesis • Designed for RT applications • Originally implemented in C4 Engine
- 16. Transvoxel overview • Based on MC Extend equivalence classes •
- 17. Basic Algorithm • • • • Subdivide grid in blocks (once) Polygonize blocks (once) Decide blocks to draw and LOD (per-frame) Draw relevant blocks (per-frame)
- 18. The LOD solution • • • Introduce “transition cells” between levels Restricted octree for rendering Identify neighbors and draw correct transitions
- 19. Transition cells
- 21. Texturing Per-voxel material Id → Texture Ids • Triplanar projection • Two sets of textures are blended for better transitions •
- 22. Parallelism • Embarrassingly parallel • Relatively easy for GPGPU • Parts can be re-computed independently
- 23. DEMO
- 24. Dual contouring • • Solves the sharp edges Works on hermite data sets VoxelFarm engine by Voxel Farm Inc.
- 25. Comparison • MC-based algorithms faster require less memory • Dual algorithms support more geometries
- 26. Potential headaches • Memory consumption • Correct implementations sometimes tricky • Texturing