Projects

This page highlights selected research and software projects. My projects are typically motivated by a combination of scientific curiosity, practical utility, and long-term research potential.

1. Weakly Supervised Pneumonia Localization

A medical imaging research project focused on detecting and localizing pneumonia-relevant regions from chest X-rays using only image-level supervision. The project investigates Grad-CAM-based interpretation, localization behavior, and the strengths and limitations of weak supervision in clinical imaging.

Focus areas: weak supervision, X-ray analysis, explainability, localization

2. Uncertainty-Aware Lung Segmentation

A segmentation project using nnU-Net v2 ensembles to study predictive confidence and model reliability in chest X-ray lung segmentation. The project emphasizes not only segmentation accuracy, but also calibration, uncertainty quantification, and robustness under dataset shift.

Focus areas: segmentation, uncertainty estimation, ensemble learning, clinical reliability

3. Temporal Video Matting Research

An ongoing research direction exploring temporal consistency, edge fidelity, and efficient modeling for video matting. The aim is to build systems that are accurate, lightweight, and capable of handling challenging temporal variations.

Focus areas: video matting, temporal learning, real-time vision, visual quality

4. Research-Driven AI Software Systems

Beyond individual research prototypes, I am interested in designing full-stack AI systems that connect model outputs with usable interfaces, reproducible workflows, and real-world deployment.

Focus areas: AI product engineering, research software, deployable systems