Camera positioning, light source positioning, and guidelines while positioning.

The input 3D point set can be entered via mouse (left-click), read from a file, or saved to a file.

Several computations needed during surface reconstruction can be performed separately on the entire set of points and then visualized:

• Nearest neighbor selection. The query point is selected by a right-click.
• PCA (Principal Components Analysis).
• MST (Minimum Spanning Tree).

Various stages of the surface reconstruction can be visualized:

• Phase 1: Tangent plane normals.
• Phase 2a: Tangent plane MST.
• Phase 2b: Consistent tangent plane normals.
• Phase 3: Signed distance function.
• Phase 4: The final reconstructed surface.

The size of tangent plane neighborhoods and the marching cube size can be adjusted for surface reconstruction.

The following screen shot shows the point set that the paper calls mechpart:

Phase 1's tangent plane normals shown with the original mechpart point set:

Phase 2a's tangent plane MST shown with the original mechpart point set:

Phase 2b's consistent tangent plane normals shown with the original mechpart point set:

Phase 3's signed distance function shown with the final reconstructed mechpart surface:

Phase 4's final reconstructed mechpart surface. I used k=15, ρ+δ=∞, and cube size = 0.02. (In DSR's Surface Reconstruction dialog, k=Tangent Plane Neighbors. ρ+δ is not adjustable; it's always ∞.)

Here's the surface for the point set that the paper calls knot. I used k=22, ρ+δ=∞, and cube size = 0.02:

Here's the surface for the point set that the paper calls cat. I used k=15, ρ+δ=∞, and cube size = 0.02:

I tried cat again with k=39, ρ+δ=∞, and cube size = 0.02. Lighting and camera position are exactly the same. Notice that some of the defects are smoothed out, but some of the expected features, such as the ears, have lost their shape a little.