Gaussian Splatting Workflows - Generating Sparse Point Cloud, Images and Corresponding Masks

Development Asset Creation
, , ,
1

Hi,

I’m able to generate sparse point clouds and colmap data relatively easily from RealityScan to put through various Gaussian Splat pipelines.

I’d like to be able to make use of the object masking feature to improve my splatting results and have been able to generate and export AI masks however these are different dimensions to the images exported during colmap export where the RealityScan distorts and in other instances even crops input data.

What is the best process to ensure that all my images exported via ColMap and from the AI mask export are consistent?

Many thanks

2

Hello David, and thank you for your patience. To export undistorted masks that match the COLMAP images, go to Scene 3D → Tools → Export Depth & Masks and choose the same Undistortion settings you use for COLMAP (this ensures the masks are resampled/cropped to the exact dimensions of the undistorted images)

You’ll need to create a model first, since this exporter generates masks from the mesh/model.

Kind regards

MC

Here’s some doc regarding this topic:

https://rshelp.capturingreality.com/en-US/tools/mask.htm