Presentation is loading. Please wait.

Presentation is loading. Please wait.

Ravish Mehra Subodh Kumar IIT Delhi IIT Delhi

Similar presentations


Presentation on theme: "Ravish Mehra Subodh Kumar IIT Delhi IIT Delhi"— Presentation transcript:

1 Ravish Mehra Subodh Kumar IIT Delhi IIT Delhi
Accurate and Efficient Rendering of Detail using Directional Distance Maps Ravish Mehra Subodh Kumar IIT Delhi IIT Delhi Technical contributions Sphere tracing with DDM Analytical ray-mesostructure intersection Lossless distance map compression Features Handles both heightfield and non-heightfield mesostructures (any model as distance map) Accurate results, no aliasing Memory efficient

2 Mesostructures microstructures mesostructures macrostructures © ACM
What is mesostructure ? > microstucture (BRDFs) < macrostructure (displacement maps) microstructures mesostructures macrostructures © ACM

3 Mesostructures height field vs non-height field
alters shadows, occlusion, and silhouettes heightfield or non-height field alters shadows, occlusion and silhouettes © ACM

4 Mesostructure Rendering
Tessellation generates large number of micro-polygons Ray-tracing generates large number of triangles Per-pixel displacement mapping works at interactive rates provides performance-quality trade-off tessellation very slow ray-tracing too slow depends on number of primitives cost prohibitive for mesostructures people started using per-pixel displacement mapping

5 Per-Pixel Displacement Mapping
render low-resolution mesh cover all fragments (pixels) recompute color and depth of each fragment Basic idea render a low resolution mesh cover all fragments color and depth recomputed

6 Previous Work Height-field mesostructures Relief mapping and variants
[Policarpo et al. 05], [Risser et al. 05], [Tatarchuk 05] Cone mapping [Dummer 05], [Policarpo and Oliveira 07] Erosion and dilation map [Kolb and Salama 2005] Others [Oh et al. 06], [Tevs et al. 08], [Baboud et al. 06] Require 2D textures Store displacement along the surface normal Small preprocessing time But limited to height-field rendering Aliasing due to texture resolution

7 Previous Work Non height-field mesostructures Relief mapping
[Policarpo and Oliveira 06] : limited non-height field Generalized displacement mapping [Wang et al. 04]: height field and non height-field linear interpolation errors Sphere tracing [Hart 1996], [Donnelly 05]: height field and non height-field Required 3D texture memory limited Policarpo – handles limited non-height field GDM – linear interpolation errors Sphere tracing – linear interpolation errors

8 Sphere Tracing Based on distance maps (Stores minimum distance to surface from a point ) Can render heightfield and non heightfield mesostructures There exist efficient ways to compute distance maps

9 Sphere Tracing Drawbacks Slow convergence Limited accuracy
points close to surface Limited accuracy based on sampling resolution of distance maps High memory requirement distance maps stored as 3D texture Slow convergence – points close to surface, method progresses very slowly Limited accuracy – sampling resolution artifact ( ray marching terminated when distance < threshold ) Memory requirement for 3D textures hight

10 Directional Distance Map
“look forward” distance only in direction of ray important direction cone limits field of view of the ray small #of direction cones “front to back” Distance only in the direction of ray are important, means, “look forward” Limits consideration of nearest surface points to a direction cone A small number of direction cones suffice, basically you have to distinguish “front from back”

11 Comparison with distance map with directional distance map

12 Accurate intersection
conservative distance nearest distance from anywhere within a voxel Original sphere tracing distance values computed from center of voxel and stored Our technique distance values computed as nearest distance from anywhere within a voxel d = center d = minimum original sphere tracing our technique

13 Accurate intersection
Quadric surface approximation voxel with surface has value 0

14 Accurate intersection
Quadric surface approximation voxel with surface has value 0 perform ray-quadric intersection Explains the process of doing ray-quadric intersection

15 Distance map compression

16 RESULTS Faster convergence Only few DDM directions suffice
Quality improvement Performance statistic Compression efficiency

17 Results Faster convergence with distance map
with directional distance map 10 35 60 100 # of iterations

18 Results Only few direction cones suffice
Performance (in ms) of sphere tracing with distance map and directional distance map (DDM) at varying number of directions.

19 Results Quality improvement with distance map
with DDM + ray-quadric intersection

20 Results Quality improvement with distance map
with DDM + ray-quadric intersection

21 Results Quality improvement with distance map
with DDM + ray-quadric intersection

22 ST with DDM and our technique (ST with DDM and quadric)
Results Performance Performance (in ms) for Relief Mapping, Sphere Tracing (ST) with distance map, ST with DDM and our technique (ST with DDM and quadric)

23 Results Compression efficiency
Compression efficiency of directional distance map (DDM) for different models.

24 THE END


Download ppt "Ravish Mehra Subodh Kumar IIT Delhi IIT Delhi"

Similar presentations


Ads by Google