K-Material SDFs for Neural Attenuation Fields
Published in UC Santa Cruz, Baskin School of Engineering, 2026
This bachelor thesis develops K-Material Signed Distance Functions (SDFs) for neural attenuation fields, enabling automated multi-material reconstruction in sparse-view CT imaging. The work introduces a differentiable Soft Selector and GMM-based priors to eliminate manual hyperparameter tuning in tissue modeling.
Key Contributions:
- Formulated K-Material SDFs for automated multi-surface reconstruction
- Developed differentiable Soft Selector and GMM-based priors
- Evaluated on modern architectures analyzing physics-informed regularization effects
Advisor: Prof. Razvan Marinescu, BiomedAI Lab, UC Santa Cruz
Recommended citation: Shah, D.K. (2026). K-Material SDFs for Neural Attenuation Fields. Bachelor Thesis, University of California, Santa Cruz.
Download Paper