Editing setting presets

This page describes the available options when editing a custom settings preset.

Caution

Many of these settings are intended for advanced users, enabling to fine-tune the reconstruction process.

We recommend reading the setting descriptions carefully before changing them, and always being aware that adjustments to these settings might not have the intended effects. Bad settings could lead to low-quality results, high processing times without an increase in quality, or crashes due to exceeding the available memory.

When adjusting these settings, we thus recommend making one change at a time, to evaluate the result of each change on multiple recordings by comparing to previous results without the change, and to keep only changes which clearly improve the results.

Surface reconstruction

Voxel size [m]

Side length (in meters) of voxels for volumetric reconstruction. Smaller voxels allow for more detailed geometry, but increase the processing time and the memory requirements. Be aware that using smaller voxels only pays off if the cameras can actually deliver a suitable effective resolution.

Regularization strength

Determines how much the reconstructed geometry gets smoothed. A higher value results in stronger smoothing. Increasing regularization can help suppress geometry noise, but may also risk losing details. Decreasing regularization can help obtain more details at the risk of more geometry noise.

Air prior

If this value is set to zero, then ‘matter’ and ‘air’ will be treated equally when filling in areas in the reconstruction that have not been observed directly by any camera. However, if matter is filled in incorrectly in areas which in reality are air, then this is usually very undesirable, as it produces noticeable bulges in the result. For this reason, this setting value, if set to a positive value, may be used to bias the fill-in of unobserved areas towards air, rather than matter. The higher the setting value, the stronger air will be prioritized over matter in such areas.

Mesh decimation algorithm

Determines the algorithm which is used to simplify the reconstructed raw meshes (i.e., to reduce their vertex and triangle count). A high-quality and a high-performance algorithm are available. The high-quality algorithm will produce superior meshes to the high-performance algorithm, however, this likely becomes apparent only if you plan to edit the resulting meshes with mesh editing software, which will likely behave nicer on these meshes. The high-performance algorithm tends to be significantly faster.

Minimum volume of mesh component [m^3]

This setting specifies the minimum volume (in cubic meters) of mesh components: Components which are smaller than this threshold get discarded. This helps remove geometry noise. If you have a camera setup observing a small volume with high resolution, then you may want to adjust this to be able to reconstruct smaller objects.

Motion estimation

Evaluation distance threshold [m]

Determines the distance (in meters) below which two surfaces are classified as well-aligned. Decreasing this will cause more keyframes to be created, resulting in more accurate, but less temporally stable and larger results. Increasing this will cause less keyframes to be created, with the opposite effects.

Evaluation samples per cm^2

Determines the average number of motion evaluation samples taken per square centimeter of mesh surface. Usually does not need to be changed.

Texture mapping

Maximum texture distortion

Determines the maximum distortion factor that the texture mapping algorithm accepts for triangle surfaces. Larger values allow the texture mapping algorithm to create larger texture charts (reducing the number of texture seams, and possibly wasting less texture space), but cause some triangles to have lower or higher texture resolution than they should have in a fair distribution of texture area over the triangles.

Texture projection safety margin [pixels]

Defines a safety margin (in pixels) around uncertain areas where the image information will not be used for texturing. Increasing this setting may help improve texture quality by removing incorrect fringes of image projections; however, this comes at the risk of losing texture information. Watch out for homogeneous texture areas that appear in place of correct texture information if this value is increased too strongly.

Attention

Currently, this value must be set to an even number.

Maximum depth ratio (reconstructed vs. estimated)

If this setting is set to a non-negative value, then the estimated depth (in a camera image) is compared with the reconstructed depth (from the 3D model created using all cameras), and a color pixel (of that same camera device) is only used for texturing if those two values are similar for the pixel. The ratio to compare this threshold with is computed as: max(estimated_depth / reconstructed_depth, reconstructed_depth / estimated_depth). This can help reject bad pixels, improving the texture quality, at the risk of rejecting some valid pixels as well. Set this setting to -1 to disable it, or to a value larger than one to activate it.

Texturing of unobserved areas

Both the surface reconstruction and the texturing algorithms fill in surface areas that were not directly observed by any camera. However, for camera configurations that lead to a significant part of the subject’s surface area remaining unobserved (such as front-facing configurations which do not observe the back side of a subject), this fill-in will not yield plausible results, as too much information is missing. In this case, this setting may be used to reconstruct these areas as transparent.

Output configurations

Output configurations determine the number of triangles in the resulting meshes or videos. Multiple output configurations may be defined at the same time, which enables producing results with different quality / bandwidth levels in a single reconstruction process.

This can save time over running multiple reconstruction processes, as intermediate results early in the processing pipeline are shared between all outputs.

Note

The user interface currently only fully supports the use of a single output configuration at a time.

Each configured output configuration has the following settings:

Maximum triangle budget

The maximum number of triangles in a reconstructed mesh. Meshes will be decimated to reduce their triangle count below this threshold (or lower, as determined by the triangle reduction factor).

Triangle reduction factor

A factor on the original triangle count of the raw reconstructed meshes, which determines a target for mesh decimation: For example, a factor of 0.2 means that raw meshes will be decimated to 20% of their original triangle count (or to the maximum triangle budget, whichever is lower).

Encoding

Maximum keyframe interval

Defines the maximum interval between successive keyframes. A value of 1 forces each frame to be a keyframe (this is what setting the “Video stabilization” reconstruction setting to off does). More keyframes allow for faster seeking and lead to more geometrically accurate videos. However, they strongly increase the file size of the videos and reduce temporal stability.

Video texture encoding mode

Determines how the video texture is encoded: With high quality, but taking more time to encode, or with lower quality, but taking less time to encode.

Memory limits

Settings in this section influence the memory usage of the reconstruction process and limit the reconstruction sizes. Higher setting values allow for larger reconstructions, but increase the memory usage for all reconstructions with this settings preset.

Maximum number of blocks (x direction)

Defines the maximum number of blocks in volumetric reconstruction, in the x direction. Increasing this value allows for larger reconstruction volumes, but increases memory usage.

Maximum number of blocks (y direction)

Defines the maximum number of blocks in volumetric reconstruction, in the y direction. Increasing this value allows for larger reconstruction volumes, but increases memory usage.

Maximum number of blocks (z direction)

Defines the maximum number of blocks in volumetric reconstruction, in the z direction. Increasing this value allows for larger reconstruction volumes, but increases memory usage.

Maximum number of unknowns in surface reconstruction

Defines the maximum number of unknowns (estimated values) in volumetric reconstruction. Increasing this value allows for larger surface reconstructions, but increases memory usage.

Maximum number of vertices in reconstructed meshes

Defines the maximum number of vertices in reconstructed raw meshes (before mesh decimation). Increasing this value allows for larger reconstructed meshes, but increases memory usage.

Maximum number of indices in reconstructed meshes

Defines the maximum number of indices in reconstructed raw meshes (before mesh decimation). Increasing this value allows for larger reconstructed meshes, but increases memory usage.

Debugging

This section contains settings which may help diagnose issues with the reconstruction process.

Log verbosity

Determines the amount of log output during reconstruction.

Vulkan debug layers

Enabling Vulkan debug layers may help diagnose issues with the application’s usage of the Vulkan API. This will slow down the reconstruction process when enabled.