Experimental Results Original Image Constraints 3 D Wireframe

Experimental Results Original Image Constraints 3 D Wireframe Novel View Original Image Adaptive Hierarchical Transformation with Discontinuities G D I H A E F B The depth map is subdivided adaptively and represented in a pyramid. The depth at each level is computed by interpolating the depth at the previous (coarser) level and adding in the coefficients at the current level. • Previous work, [Szeliski 1990], interpolated the coarse depth map without considering discontinuities in the transformation step, slowing convergence. C • Our technique interpolates the coarse depth map only in the continuous neighborhood. For example, let w = [wi, j] be the hierarchical coefficients of z : z. E z. F z. G z. H z. I Our transformation Previous work ( z. A+z. B ) / 2 + w. E z C + w. F z D + w. G ( z. D+z. A ) / 2 + w. H ( z. E+z. G+z. H ) / 3 + w. I same ( z. B+z. C ) / 2 + w. F ( z. C+z. D ) / 2 + w. G same ( z. E+z. F+z. G +z. H ) / 4 + w. I Advantage: No constraint interference across discontinuity boundaries. Constraints 3 D Wireframe Novel View Performance Comparison of Different Solution Techniques Methods No hierarchical transformation Hierarchical transformation without continuitybased interpolation Hierarchical transformation with continuitybased interpolation Iteration 0 Iteration 200 Iteration 1200 Iteration 2500 Iteration 9500