A selection of technical projects I've worked on across 3D processing, machine learning, and computer vision.
Prior to transitioning to industry, my academic research also encompassed statistical data analysis, distributed systems, algorithm design, pattern recognition, and performance optimization for high-throughput computing environments.
Developed a deep learning model to extract ground support elements in underground mining scenes with over 90% recall rate. Managed training datasets and set up distributed training infrastructure on AWS.
Built algorithms to detect geometry structures and clean point clouds using Concave Hull algorithm to remove inner surfaces and occluded objects. Reconstructed surface meshes using Poisson reconstruction and implemented tracking to monitor surface changes over time intervals.
Developed interactive tools for creating mesh cutting contours and planes. Implemented refinement methods including smoothing and remeshing to prepare meshes for 3D printing.
Trained computer vision models to detect fraudulent identity documents with only 2% false positive rate, significantly improving customer onboarding experience. Implemented text detection and OCR for automated information extraction.
Applied state-of-the-art speech recognition models to noisy, radio-transmitted voice samples, achieving sub-20% word error rate in challenging acoustic conditions.
Did a demo project of multi-modal sensor fusion combining camera images and LiDAR point clouds on the nuScenes autonomous driving dataset. The fusion results are used for scene understanding and path planning.