WEEK 03

Align Images

I ran the align images process under the Alignment tab to generate camera positions and a sparse point cloud of the scan data.

I adjusted alignment settings where necessary by:

• Setting feature detection to High if not already the default

• Removing problematic images if alignment failed

III. Mesh Generation and Cleanup

Create Initial Mesh

I generated a Normal Detail mesh to inspect alignment quality and identify artifacts.

Filter and Refine

I used the Lasso tool to select and remove unwanted geometry using Filter Selection.

Clean Model and Close Holes ensure the quality of the mesh is maintained and remove any topological defects.

For further refinement, I exported selections as masks and iterated in Nuke to improve tassel removal. I created additional garbage mattes from filtered selections in RealityScan to isolate the seal body and remove remaining background and tassel elements.

This established an iterative workflow:

RealityScan → Mask refinement (Nuke) → RealityScan,

which I used to improve results for thin or complex geometry, reflective and subsurface surface details, and background spill during reconstruction.

For consistent results, I restarted the project after updating masks and folder structure.

IV. High-Detail Reconstruction

Once the alignment and masks were verified, I initiated the High Detail reconstruction. Depending on image count and hardware, this process took between 30 minutes and 2 hours.

After the first reconstruction, I generated multiple iterations to evaluate mesh quality:

• Version 01: Underexposed processed scan images using XPOL geometry

• Version 02: Identified underexposed images, applied exposure compensation, and regenerated the mesh using geometry and mask layers

• Version 03: Tested PPOL images in the _geometry folder after exporting 8-bit PNGs

• Version 04: Tested combined PPOL and XPOL images for the geometry layer

Version 01

Findings

I observed that:

• PPOL-only reconstruction produced unstable geometry due to varying specular reflectivity and subsurface variation (version 03)

• Combined PPOL + XPOL improved stability but reduced fine detail (version 04)

• XPOL-only reconstruction preserved the highest level of surface detail (version 02)

This confirmed the following principle:

• XPOL → best for geometry (diffuse, stable features)

• PPOL → best for texture/specular data

Mixing both reduced micro-detail fidelity due to inconsistent feature responses.

The final result was Version 2 (XPOL-based), which produced the most reliable mesh.

V. Mesh Optimization (Master Creation)

High-detail reconstruction produced dense geometry, often ranging between 50M–300M polygons. At these densities, the viewport may default to a point cloud display to maintain performance.

The Jade Seal reconstruction resulted in approximately 27M polygons.

1. Evaluate

I inspected the mesh for surface quality and completeness using tools such as Check Topology, Integrity, and Quality Analysis in the Mesh and Color tab.

2. Simplify

Once the scan was successful, I used the Simplify tool to reduce mesh density while preserving key detail.

3. Define Master Mesh

I defined the simplified mesh as the Master High-Poly, which will be used for all subsequent baking and cleanup.

VI. UV Unwrapping and Texture Generation

Before baking, I defined the texture layers and unwrapped the model.

UV Unwrap

Under the Mesh Model tab, I clicked Unwrap to unwrap the mesh to a single texture.

For a single texture setup, I set the following:

• Unwrapping style → Maximal Texture Count

• Maximal texture count → 1

Texture Generation

Under the Mesh Model tab, I clicked Texture to begin the texturing process.

In the Scene 3D > View tab, I reviewed the master mesh using different display modes (Vertices, Solid, or Sweet).

Note: Models larger than 40 million triangles display as a dense colored point cloud.

VII. Final Export

Once the textures and UVs were done generating, I exported the 27M master mesh and its associated textures for further cleanup and asset processing in Houdini.

Result

I generated a clean high-resolution reconstruction of the Jade Seal with the tassel successfully removed. This established a stable workflow combining cross-polarized geometry reconstruction with iterative mask refinement, producing an initial draft of the cleanup master mesh for further cleanup and asset processing.