310 likes | 436 Views
An Incremental Weighted Least Squares Approach To Surface Light Fields. Greg Coombe Anselmo Lastra. Image-Based Rendering. Generate new views of a scene from existing views Sample appearance from physical world. Lightfields [Levoy96], Lumigraphs [Gortler96]. …. Surface Lightfields.
E N D
An Incremental Weighted Least Squares Approach To Surface Light Fields Greg Coombe Anselmo Lastra
Image-Based Rendering • Generate new views of a scene from existing views • Sample appearance from physical world Lightfields [Levoy96], Lumigraphs [Gortler96] …
Surface Lightfields • A surface light field [Wood00] represents the appearance of a model with known geometry and static lighting (θ,Φ) surface position (u,v) viewing direction
Surface Lightfields f1(Θ, Φ) f2(Θ, Φ) f3(Θ, Φ) f4(Θ, Φ)
SLF - Batch Process Surface Lightfield Construction Renderer disk … [Chen02, Hillesland03]
SLF - Online Process Incremental Surface Lightfield Construction Renderer 1024x768 @ 15fps [Coombe05]
Scattered Data Approximation • [Coombe05] required resampling to grid • Samples are at arbitrary locations in domain due to geometry and camera ? ?
Scattered Data Approximation • Scattered Data Approximation in lightfields • Unstructured Lightfields [Buehler01] • Tesselation of pure lightfield • Polynomial Texture Maps [Malzbender01] • Fit polynomials to set of images • Radial Basis Functions [Zickler05] • Interpolate sparse reflectance data
Outline • A 2D scattered data approximation • Fast incremental construction • Realtime evaluation
Our Approach • Represent the surface lightfield using Weighted Least Squares approximation • Modify WLS for the incremental framework • Adaptive and Hierarchical • CPU/GPU implementation • Process each image in about 1-2 seconds • Real-time rendering
Bust model, 75 training images
Least Squares Fitting • Find the “best” polynomial approximation to the input samples • “best” means minimizes sum of squared differences • the coefficients are determined by solving a linear system input samples reconstruction
Least Squares reconstructed function input samples domain
Weighted Least Squares • Problem: LS is a global approximation • Solution: Divide domain into multiple LS approximations, and combine to get global approximation • Use a set of low-degree polynomials • Non-linear blending (Partition of Unity) • Good discussion in Scattered Data Approximation, Holgar Wendland
Weighted Least Squares reconstructed function polynomial approximations input samples domains centers
Weighted Least Squares Θ x x x x Φ
Outline • A 2D scattered data approximation • Fast incremental construction • Realtime evaluation
Incremental WLS • Feedback is important in SLF construction • As each image is captured, it must be incorporated into the representation • How do we determine the size of each domain?
Adaptive Construction • Start out with large domains • Adaptive shrink as more points arrive x x x x x x x x
Hierarchical Construction • Start out with a single domain • Subdivide as more points arrive (quadtree) x x x x x x x x x x x
Hierarchical Construction, First 10 images
Hierarchical and Adaptive • Hierarchical • 1x1 Polynomial Texture Mapping • Fast at first, slows down as refines • # of domains is a power of 2 • Adaptive • Slow at first, speeds up as domains shrink • Can handle arbitrary # of domains
Outline • A 2D scattered data approximation • Fast incremental construction • Realtime evaluation
Implementation Pose Estimation Visibility / Reprojection Incremental Weighted Least Squares 1024x768 @ 15fps GPU Renderer
Θ Θ Φ Φ Θ Φ Θ Φ GPU Implementation
Results 4K patches 29K patches side by side, 14K patches
Pitcher model, 65 images
Performance • ~ 0.5 - 2 seconds per image for hierarchical construction • 0.5s for 4K bust model • 2s for 30K pitcher model • 95% is Least Squares Fitting • Adaptive is 2-3x more expensive • Rendering is 60fps or more
Conclusion • Represent the surface lightfield using Weighted Least Squares approximation • Modify WLS for the incremental framework • Adaptive and Hierarchical • CPU/GPU implementation • Process each image in about 1-2 seconds • Real-time rendering
Future Work • Since construction is dominated by LS fitting, implement on GPU • Extend to surface reflectance (4D) • Change basis functions • Order-of-magnitude more data
Thanks • NVIDIA Graduate Student Fellowship • UNC Graphics Group • National Science Foundation • Questions?