1. Virtual Light Field Group vlfproject@cs.ucl.ac.uk University College London GR/R13685/01 Research funded by: Propagating the VLF - Problems and Solutions II Insu Yu i.yu@cs.ucl.ac.uk

2. VLF Project Overview Specular Transfer – Determination of reflected Direction – Forward/Backward Transfer – Discontinuity and Resampling Caustics – What is caustics – Holes & Jittering Propagation Issues – Progressive propagation – Effect of Parameter Adjustment

3. VLF Project Specular Reflection What is Ideal specular reflection model ? – All the incoming Energy (L) is reflected off the direction(R) of a surface – Incidence angle ( θ i ) equals the angle of specular reflection ( θ r ) – A Mirror Reflection How can we adapt this model on VLF ?

4. VLF Project Determination of reflected direction Discrete ‘Uniformly Sampled Directions’ – Due to the Representation of Finite Directions, an Ideal reflected PSF direction can hardly be found PSF Directions (Quadrant) VLF Specular Reflection

5. VLF Project Nearest Vs Tri-linear Directions Nearest direction – Select closest direction to the reflected direction – Works better on High Resolution Directions Finding Nearest PSF Direction

6. VLF Project Nearest Vs Tri-linear Directions (Cont’) Tri-linear – Select Three PSF direction which enclose the reflected Direction – Transfer Incoming Energy to three Directions according to barycentric weights of each direction – Glossy not perfect specular – Tri-linear interpolated transfer introduce excessive blurring Tri-linear Transfer

7. VLF Project Forward Vs Backward Mapping Forward mapping – Diffuse to Specular Transfer – Shooting Energy from Sender to Receiver – Miss-alignment of source to destination leads holes Forward Mapping Holes

8. VLF Project Forward Vs Backward Mapping (Cont’) Backward mapping – Trace a ray in Reflected Direction in back order – One-to-One mapping – UV uniform subdivision on receiver planes – Need to use bilinear interpolation scheme to pull the radiance values Backward Mapping

9. VLF Project Discontinuities & Resampling Specular Reflection is not jittered – Diffuse Surfaces are jittered multiple time – Due to Directional Energy Transfer – URM is updated only once where Diffuse to Specular Transfer occurs Blocky discontinuities appear despite Tri- linear transfer Resample all TRM at end of propagation by backwards ray tracing Resampling

10. VLF Project Relation to Directions Accuracy of Specular Transfer is proportional to the number of directions Light Field Ray Tracing Specular Reflection (ES*D)

11. VLF Project Caustics

12. VLF Project Caustics Caustics comes free as in Specular to Diffuse Transfer

13. VLF Project Specular to Diffuse Transfer Caustics present where Diffuse surface gathers energy from specular senders

14. VLF Project Holes and Jittering Holes Artefact – Due to representation of Discretisation Directions – Transfer from small reflector to large receiver – Long distance between reflector and receiver Increase sampling density of directions More jittered samples for caustic

15. VLF Project Filtering Filtering to achieve accuracy and avoid aliasing Caustics on Diffuse map can be excessively Blurred & introduce light leaking – Require Adjustment of Gaussian Kernel Sigma Possible to have higher resolution diffuse maps on caustic receivers Can exploit additional map for caustics

16. VLF Project Propagation Propagation Process over various iterations

17. VLF Project Progressive Propagation Progressive propagation framework – Estimating unshot radiance – Selecting shooting sources & managing swapping of maps – Purging of unshot energy – Zero-energy surfaces are never senders Other issue – Capping of scenes

18. VLF Project Scalability Test ( Effect of parameters) Various Polygons Various PSF Directions Various PSF Size & Tile Size Propagation time and on memory Dual Xeon 1.7Ghz

19. VLF Project Effect of parameters (Polygons) Propagation time varies quadratically with the number of polygons (513 Direction, 8x8 Tiles, 64x64 Cells) The memory grows linearly with the number of polygons TEST Scene – One Emitter – polygons 224 to 1736 – 5:1 ratio of diffuse to specular surfaces

20. VLF Project Effect of parameters (Directions) Increasing PSF Directions – Pros The greater accuracy is achieved Less jittering is necessary (overcome missing holes) – Cons More memory usage Longer Propagation Time Linear Relationship between the number of direction and propagation time/memory Office Test Scene

21. VLF Project Effect of parameters (PSF/Tile Resolution ) Size of Tile/PSF resolution determine the speed of propagation & rendering Increasing Resolution results in faster intersection searching but more memory & propagation time

22. VLF Project Question ?