EFFICIENT SPATIOTEMPORAL RESAMPLING USING PROBABILITY DENSITY FUNCTION SIMILARITY

20240193847

18/076496

December 07, 2022

A processor shares path tracing data across sampling locations to amortize computations across space and time. The processor maps a group of sampling locations of a frame that are adjacent to each other to a reservoir. Each reservoir is associated with a ray that intersects subsets of path space such as a pixel. The processor resamples the reservoirs based on a similarity of probability density functions (PDFs) between pixels to select a set of samples mapped to the reservoir. The processor then performs resampling of the selected set of samples to obtain a representative light sample to determine a value for each pixel and renders the frame based on the values of the pixels.

1. A method comprising: comparing a first probability density function (PDF) for a first pixel and a second PDF for a second pixel to obtain a similarity of the first PDF and the second PDF; reusing samples from the first pixel for resampling rays to select a set of samples comprising rays that intersect subsets of path space at the second pixel based on the similarity; and rendering a first frame based on the selected set of samples.

2. The method of claim 1, wherein comparing comprises: approximating the first PDF with a first von Mises-Fisher (vMF) distribution; approximating the second PDF with a second vMF distribution; and computing a similarity of the first vMF distribution and the second vMF distribution.

3. The method of claim 2, wherein the first vMF distribution is based on a weighted average of past sample directions for the first pixel and the second vMF distribution is based on a weighted average of past sample directions for the second pixel.

4. The method of claim 1, wherein reusing samples comprises reducing an amount of spatial reuse of samples from the first pixel based on the similarity.

5. The method of claim 4, wherein reusing samples further comprises: applying weights to the samples from the first pixel for resampled importance sampling based on the similarity.

6. The method of claim 5, wherein reusing samples further comprises: computing an average direction of rays that intersect subsets of path space at the second pixel based on a sum of weighted directions divided by a sum of weights.

7. The method of claim 1, wherein the first pixel spatially neighbors the second pixel.

8. A non-transitory computer readable medium embodying a set of executable instructions, the set of executable instructions to manipulate at least one processor to: compare a first probability density function (PDF) for a first pixel and a second PDF for a second pixel to obtain a similarity of the first PDF and the second PDF; reuse samples from the first pixel for resampling rays to select a set of samples comprising rays that intersect subsets of path space at the second pixel based on the similarity; and render a first frame based on the selected set of samples.

9. The non-transitory computer readable medium of claim 8, wherein the at least one processor is to: approximate the first PDF with a first von Mises-Fisher (vMF) distribution; approximate the second PDF with a second vMF distribution; and compute a similarity of the first vMF distribution and the second vMF distribution.

10. The non-transitory computer readable medium of claim 9, wherein the first vMF distribution is based on a weighted average of past sample directions for the first pixel and the second vMF distribution is based on a weighted average of past sample directions for the second pixel.

11. The non-transitory computer readable medium of claim 8, wherein the at least one processor is to: reduce an amount of spatial reuse of samples from the first pixel based on the similarity.

12. The non-transitory computer readable medium of claim 11, wherein the at least one processor is to: apply weights to the samples from the first pixel for resampled importance sampling based on the similarity.

13. The non-transitory computer readable medium of claim 12, wherein the at least one processor is to: compute an average direction of rays that intersect subsets of path space at the second pixel based on a sum of weighted directions divided by a sum of weights.

14. The non-transitory computer readable medium of claim 8, wherein the first pixel spatially neighbors the second pixel.

15. A device comprising: a memory to store a plurality of light sources having rays that intersect a set of sampling locations at a first pixel; and a processor coupled to the memory to: compare a first probability density function (PDF) for a first pixel and a second PDF for a second pixel to obtain a similarity of the first PDF and the second PDF; reuse samples from the first pixel for resampling rays to select a set of samples comprising rays that intersect subsets of path space at the second pixel based on the similarity; and render a first frame based on the selected set of samples.

16. The device of claim 15, wherein the processor is to: approximate the first PDF with a first von Mises-Fisher (vMF) distribution; approximate the second PDF with a second vMF distribution; and compute a similarity of the first vMF distribution and the second VMF distribution.

17. The device of claim 16, wherein the first vMF distribution is based on a weighted average of past sample directions for the first pixel and the second vMF distribution is based on a weighted average of past sample directions for the second pixel.

18. The device of claim 15, wherein the processor is to: reduce an amount of spatial reuse of samples from the first pixel based on the similarity.

19. The processor of claim 18, wherein the processor is to: apply weights to the samples from the first pixel for resampled importance sampling based on the similarity.

20. The device of claim 18, wherein the processor is to: compute an average direction of rays that intersect subsets of path space at the second pixel based on a sum of weighted directions divided by a sum of weights.

06; 06

edspap.20240193847