Computer Vision Research · ASU Dream Lab

01 / OVERVIEW

Context & problem

Built Python/PyTorch data pipelines and repeatable research loops for 3D reconstruction, adapting deep-learning methods to NASA lunar datasets.

Limited viewpoints can leave scene geometry underconstrained, producing incomplete representations and reconstruction errors. The research explored synthetic training data and TSDF-augmented strategies to investigate model behavior under sparse views.

02 / SYSTEM ARCHITECTURE

How the pieces connect.

03 / INDIVIDUAL CONTRIBUTION

What I worked on.

• Built data pipelines for repeatable computer-vision experiments.
• Designed synthetic-data generation for occupancy networks.
• Investigated few-view Neural Radiance Field failure modes.
• Explored TSDF-augmented training and evaluated reconstruction behavior.

04 / IMPLEMENTATION

Engineering details.

Limited viewpoints can leave scene geometry underconstrained, producing incomplete representations and reconstruction errors. The research explored synthetic training data and TSDF-augmented strategies to investigate model behavior under sparse views.

• Prepared dataset and experiment workflows in Python/PyTorch.
• Used synthetic data to explore occupancy-network training.
• Studied few-view NeRF behavior and TSDF augmentation.

05 / RESULTS & VALIDATION

What the work supports.

• Contributed to experiment reproduction and adaptation on lunar data.
• No accuracy or comparative metric is claimed.

Validation details are limited to evidence stated in the resume; no unreported datasets, accuracy claims, or benchmark comparisons are added.