Drezil/masterarbeit
1
1
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

Update Readme.md

Browse Source
This commit is contained in:
Nicole Dresselhaus 2024-07-11 15:49:05 +00:00
parent 9cf8050324
commit e0f0c5258f
1 changed files with 13 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
Readme.md
View File

@ -1,10 +1,13 @@
<p><strong>Title</strong>: Evaluating Evolvability Criteria for Deformable Objects Using Free-Form Deformation</p> **Title**: Evaluating the Performance of Randomized FFD Control Grids for Mesh Deformation in Computational Geometry and Design Optimization
<p><strong>Short summary</strong>: This thesis investigates the predictive power of evolvability criteria—variability, regularity, and improvement potential—in optimizing deformable objects using free-form deformation (FFD). We replicate Richter et al.'s study on radial basis functions (RBF) with FFD to understand how well these criteria forecast the optimization quality. Our results indicate that variability and improvement potential are reliable predictors for 3D object fitting, while regularity's correlation is inconsistent across scenarios.</p>
<p><strong>Methodology</strong>: We used evolutionary algorithms to optimize geometric objects represented through FFD grids. The study involved creating various control-grid resolutions and deformations in both one-dimensional (plane approximation) and three-dimensional (mesh fitting) settings. Evolvability criteria were calculated for each grid setup, correlating them with the convergence speed of the evolutionary algorithm and the quality of object fit.</p> **Short summary**: This paper explores the effectiveness of various evolvability criteria, namely regularity, variability, and improvement potential, when applied to randomized Free Form Deformations (FFD) control grids. Through empirical analysis using both 1-dimensional and 3-dimensional deformation scenarios, it assesses how these metrics correlate with the quality of fitness outcomes from evolutionary optimization processes in computational geometry tasks such as function approximation and mesh fitting.
<p><strong>Results</strong>:</p>
<ul> **Methodology**: The study utilizes Free Form Deformations (FFD) to parameterize complex shapes for manipulation via control grids, which are then evolved using a CMA-ES algorithm. Two scenarios were considered – 1D function approximation with known analytical solutions and 3D mesh fitting without an exact solution, employing Spearman's rank correlation coefficient to evaluate the relationship between evolvability criteria and fitness results.
<li><em>Main takeaway</em>: Variability is a strong predictor of optimization success in both 1D and 3D scenarios; improvement potential also shows significant correlation to fitting error across all grid resolutions.</li>
<li><em>Strengths</em>: The study successfully replicates the findings for variability and improvement potential as robust indicators for deformable object quality, similar to Richter et al.'s results using RBF.</li> **Results**:
<li><em>Weaknesses</em>: Regularity's effectiveness as a predictor is inconsistent; it correlates well in some scenarios but not in others or shows anti-correlation with convergence speed.</li> - *Main takeaway*: The paper finds that variability and improvement potential are robust predictors of deformation quality in FFD control grids. Variability showed a very strong, significant positive correlation with fitting error for both scenarios (1D and 3D), suggesting its usefulness as an indicator of design space exploration capability. Improvement potential also displayed a very strong, significant negative correlation with fitting error across all test cases.
<li><em>Open questions</em>: Further research is needed to refine the regularity criterion and explore direct FFD manipulations like DM-FFD for better optimization quality predictions.</li> - *Strengths*: The paper successfully demonstrates the utility of evolvability metrics in predicting fitness outcomes for FFD control grid configurations without needing an exact solution to optimization problems. It further identifies improvement potential as a sensitive measure that can estimate deformation quality even with varying gradient information.
</ul> - *Weaknesses*: The regularity metric did not consistently correlate with convergence speed or fitting error, indicating its limited predictive power in FFD contexts. This may be due to the presence of control points contributing insignificantly to mesh parameterization, affecting the condition number of the deformation matrix and thus misrepresenting local effects on fitness outcomes.
- *Open questions*: The paper raises the question of how to refine regularity as an evolvability criterion for FFD grids, suggesting that incorporating all singular values might improve its effectiveness in capturing local deformation characteristics. It also suggests further investigation into direct manipulation methods (like DM-FFD) and their interaction with evolvability criteria.
Note: This summary was automagically generated using a good™ prompt on microsofts phi3:14b-medium-128k-f16 LLM.