Semester Project

Code base for the semster thesis Local 3D Mesh Refinement via Gaussian Mixture Modelling.

Setup

To set this pipeline up you need to configure your virtual environment as written in install/requirements.txt:

pip install -r install/requirements.txt

Further you need to build and install the repository directGMM (branch:inital testing) and pymesh (follow the install instructions on the website).

Overview

All the building blocks of the pipeline are gathered in the lib directory. The scripts on the top level represent various applications of the introduced tools.

Scripts

• automated_evaluation: Scripted evaluation for a given set of parameters. main takes a parameters script and returns a box plot of the evaluation. The algorithm takes a mesh, corrupts it and then does the refinement step.

• final_evaluation: Operates automated_evaluation and feeds a set of parameters. Creates the plots shown in the reports evaluation.

• gmm_comparison: Intends to do analyze the t-test and shows the returned scores.

• mesh_editor: Helper script to edit and pertube meshes.

• pipeline_dummy: Example implementation of how to process a mesh and a pointcloud.

• presentation_plots: Same as automated evaluation but with fixed view points and sensor positions.

• two_mesh_evaluation: Mesh refinement for a ground true and a seperate corrupted mesh.

• viusalize_mesh: Helper script to select view points and camera positions.

Pipeline components

All these libraries are found in lib

• evaluation: Contains all the evaluation score logic
• gmm_generation: All gmm transforms are implemented here.
• loader: Handles all interfaces to load meshes, pointclouds, gmms
• merge: Provides the T test with its helpers and a simple placeholder merge.
• registration: Finds the regirstation of the pointcloud with respect to the given mesh. (Used in pipeline_dummy)
• visualization: Whole range of plotting facilities for meshes, pointclouds and gmms. Also creates match matrices and box plots.