260 likes | 444 Views
Adaptive Progressive Photon Mapping. Adaptive PPM. Original PPM. Anton S. Kaplanyan Karlsruhe Institute of Technology, Germany. Progressive Photon Mapping in Essence. Pixel estimate using eye and light subpaths Generate full path by joining subpaths. Photon radiance. Eye subpath
E N D
Adaptive Progressive Photon Mapping • Adaptive PPM • Original PPM Anton S. Kaplanyan Karlsruhe Institute of Technology, Germany
Progressive Photon Mapping in Essence Pixel estimate using eye and light subpaths Generate full path by joining subpaths Photon radiance Eye subpath importance Kernel-regularized connection of subpaths
Reformulation of Photon Mapping PPM = recursive (online) estimator [Yamato71] Rearrange the sum to see that Kernel estimation Path contribution
Radius Shrinkage Shrink radius (bandwidth) for th photon map User-defined parameters and Problem: Optimal value of and areunknown Usually globally constant / k-NN defined
User Parameters Example Box scene (reference)
User Parameters Example Larger Difference image Larger 𝛼
Optimal Convergence Rate • Variance and bias depend on [KZ11] Optimal rate is with • Asymptotic convergence Unbiased Monte Carlo is faster:
Convergence Rate of Kernel Estimation Convergence rate for dimensions Suffers from curse of dimensionality Adding a dimension reduces the rate! Shutter time kernel estimation – not recommended Wavelength kernel estimation – not recommended Volumetric photon mapping
Adaptive Bandwidth Selection might not yield minimal Minimize with respect to • Achieve variance ↔ bias tradeoff • Select optimal using past samples
Estimation Error • Mean Squared Error [Hachisuka et al. 2010]
Estimation Error • Variance is two-fold: • Path measurement contribution • Kernel estimation
Estimation Error • Measurement variance is higher
Estimation Error So, MSE has noise(path variance) and bias Variance Bias
Adaptive Bandwidth Selection • Both variance and bias depend on • Where is a pixel Laplacian Laplacian is unknown
Estimating Pixel Laplacian consists of Laplacians at all shading points • Weighted per-vertex Laplacians
Estimating Per-Vertex Laplacian Estimate per-vertex Laplacian at a point Recursive finite differences [Ngen11] • Yet another recursive estimator • Another shrinking bandwidth • Robust estimation on discontinuities
Adaptive Bandwidth Selection Estimate all unknowns Path variance Pixel Laplacian Minimize MSE as MSE(r) Lower initial error • Keeps noise-bias balance • Data-driven bandwidth selector
Results Progressive Photon Mapping Adaptive PPM 20 seconds!
Results Progressive Photon Mapping Adaptive PPM 3 seconds!
Conclusion Optimal asymptotic convergence rate • Asymptotically slower than unbiased methods • Not always optimal in finite time Adaptive bandwidth selection • Based on previous samples • Balances variance-bias • Speeds up convergence • Attractive for interactive preview